chore: release v3.97.4

* chore(dataset-run-items): add env variable for trace creation

* chore: push

* chore: push

.nvmrc

@@ -1 +1 @@
-v20
+v20.19.2

fern/apis/server/definition/annotation-queues.yml

@@ -22,6 +22,13 @@ service:
             docs: limit of items per page
       response: PaginatedAnnotationQueues
 
+    createQueue:
+      docs: Create an annotation queue
+      method: POST
+      path: /annotation-queues
+      request: CreateAnnotationQueueRequest
+      response: AnnotationQueue
+
     getQueue:
       docs: Get an annotation queue by ID
       method: GET
@@ -168,6 +175,12 @@ types:
       data: list<AnnotationQueueItem>
       meta: pagination.MetaResponse
 
+  CreateAnnotationQueueRequest:
+    properties:
+      name: string
+      description: optional<string>
+      scoreConfigIds: list<string>
+
   CreateAnnotationQueueItemRequest:
     properties:
       objectId: string

package.json

@@ -1,6 +1,6 @@
 {
   "name": "langfuse",
-  "version": "3.96.2",
+  "version": "3.97.4",
   "author": "engineering@langfuse.com",
   "license": "MIT",
   "private": true,

packages/shared/src/env.ts

@@ -158,6 +158,12 @@ const EnvSchema = z.object({
   LANGFUSE_EXPERIMENT_INSERT_INTO_AGGREGATING_MERGE_TREES: z
     .enum(["true", "false"])
     .default("false"),
+  LANGFUSE_EXPERIMENT_WHITELISTED_AMT_TABLES: z
+    .string()
+    .optional()
+    .transform((s) =>
+      s ? s.split(",").map((s) => s.toLowerCase().trim()) : [],
+    ),
   LANGFUSE_INGESTION_PROCESSING_SAMPLED_PROJECTS: z
     .string()
     .optional()
@@ -190,10 +196,10 @@ const EnvSchema = z.object({
         return new Map<string, number>();
       }
     }),
-
   SLACK_CLIENT_ID: z.string().optional(),
   SLACK_CLIENT_SECRET: z.string().optional(),
   SLACK_STATE_SECRET: z.string().optional(),
+  HTTPS_PROXY: z.string().optional(),
 
   LANGFUSE_SERVER_SIDE_IO_CHAR_LIMIT: z.coerce
     .number()

packages/shared/src/features/evals/utilities.ts

@@ -27,16 +27,42 @@ export const parseUnknownToString = (value: unknown): string => {
   return String(value);
 };
 
+/**
+ * Recursively parses JSON strings that may have been encoded multiple times.
+ * This handles cases where data has been JSON.stringify'd multiple times.
+ *
+ * @param value - The potentially multi-encoded JSON string
+ * @returns The final parsed object or the original value if parsing fails
+ */
+function parseMultiEncodedJson(value: unknown): unknown {
+  if (typeof value !== "string") {
+    return value;
+  }
+
+  try {
+    const parsed = JSON.parse(value);
+
+    // If result is still a string, it might be double-encoded - recurse
+    if (typeof parsed === "string") {
+      return parseMultiEncodedJson(parsed);
+    }
+
+    return parsed;
+  } catch {
+    // If parsing fails, return original value
+    return value;
+  }
+}
+
 function parseJsonDefault(selectedColumn: unknown, jsonSelector: string) {
+  // selectedColumn should already be preprocessed by preprocessObjectWithJsonFields
+  // so we can directly use it with JSONPath
   const result = JSONPath({
     path: jsonSelector,
-    json:
-      typeof selectedColumn === "string"
-        ? JSON.parse(selectedColumn)
-        : selectedColumn,
+    json: selectedColumn as any, // JSONPath accepts unknown but types are strict
   });
 
-  return result.length > 0 ? result[0] : undefined;
+  return Array.isArray(result) && result.length > 0 ? result[0] : undefined;
 }
 
 export function extractValueFromObject(
@@ -44,7 +70,13 @@ export function extractValueFromObject(
   mapping: z.infer<typeof variableMapping>,
   parseJson?: (selectedColumn: unknown, jsonSelector: string) => unknown, // eslint-disable-line no-unused-vars
 ): { value: string; error: Error | null } {
-  const selectedColumn = obj[mapping.selectedColumnId];
+  let selectedColumn = obj[mapping.selectedColumnId];
+
+  // Simple preprocessing: attempt to parse to valid JSON object
+  if (typeof selectedColumn === "string") {
+    selectedColumn = parseMultiEncodedJson(selectedColumn);
+  }
+
   const jsonParser = parseJson || parseJsonDefault;
 
   let jsonSelectedColumn;

packages/shared/src/server/data-deletion/ingestionFileDeletion.ts

@@ -88,7 +88,7 @@ async function removeIngestionEventsFromS3AndDeleteClickhouseRefs(p: {
   );
   await softDeleteInClickhouse(blobStorageRefs);
   logger.info(
-    `Deleted batch ${batch} of size ${blobStorageRefs.length} for ${projectId} of deleting s3 refs`,
+    `Deleted last batch ${batch} of size ${blobStorageRefs.length} for ${projectId} of deleting s3 refs`,
   );
 }
 

packages/shared/src/server/dataset-run-items/datasetExecution.ts

@@ -1,5 +1,4 @@
 import { env } from "../../env";
-import { logger } from "../../server/logger";
 import {
   DatasetRunItemsExecutionStrategy,
   DatasetRunItemsOperationType,
@@ -48,23 +47,8 @@ export async function executeWithDatasetRunItemsStrategy<TInput, TOutput>({
   if (operationType === DatasetRunItemsOperationType.WRITE) {
     // For write operations, implement dual-write strategy
     if (strategy.shouldWriteToClickHouse) {
-      // Dual-write phase: write to both databases
-      const postgresResult = await postgresExecution(input);
-
-      try {
-        await clickhouseExecution(input);
-        logger.debug("Successfully wrote to both PostgreSQL and ClickHouse", {
-          operation: `dataset_run_items_${operationType}`,
-        });
-      } catch (error) {
-        logger.error("ClickHouse write failed during dual-write phase", {
-          error: error instanceof Error ? error.message : String(error),
-          operation: `dataset_run_items_${operationType}`,
-        });
-        // Continue with PostgreSQL result since it succeeded
-      }
-
-      return postgresResult;
+      // Write only to ClickHouse
+      return await clickhouseExecution(input);
     } else {
       // Write only to PostgreSQL
       return await postgresExecution(input);
@@ -74,20 +58,10 @@ export async function executeWithDatasetRunItemsStrategy<TInput, TOutput>({
     const shouldExecuteClickhouse = strategy.shouldReadFromClickHouse;
 
     if (shouldExecuteClickhouse) {
-      try {
-        return await clickhouseExecution(input);
-      } catch (error) {
-        logger.error(
-          "ClickHouse execution failed, falling back to PostgreSQL",
-          {
-            error: error instanceof Error ? error.message : String(error),
-            operation: `dataset_run_items_${operationType}`,
-          },
-        );
-        // Fallback to PostgreSQL for reliability
-        return await postgresExecution(input);
-      }
+      // Read from ClickHouse
+      return await clickhouseExecution(input);
     } else {
+      // Read from PostgreSQL
       return await postgresExecution(input);
     }
   }

packages/shared/src/server/index.ts

@@ -14,6 +14,7 @@ export * from "./llm/fetchLLMCompletion";
 export * from "./llm/utils";
 export * from "./llm/types";
 export * from "./llm/compileChatMessages";
+export * from "./llm/testModelCall";
 export * from "./utils/DatabaseReadStream";
 export * from "./utils/transforms";
 export * from "./clickhouse/client";

packages/shared/src/server/ingestion/modelMatch.ts

@@ -16,6 +16,8 @@ export type ModelMatchProps = {
   model: string;
 };
 
+const MODEL_MATCH_CACHE_LOCKED_KEY = "LOCK:model-match-clear";
+
 export async function findModel(p: ModelMatchProps): Promise<Model | null> {
   return instrumentAsync(
     {
@@ -78,6 +80,14 @@ const getModelFromRedis = async (
   }
 
   try {
+    if (await isModelMatchCacheLocked()) {
+      logger.info(
+        "Model match cache is locked. Skipping model lookup from Redis.",
+      );
+
+      return null;
+    }
+
     const key = getRedisModelKey(p);
     const redisModel = await redis?.get(key);
     if (redisModel) {
@@ -241,6 +251,70 @@ export async function clearModelCacheForProject(
       );
     }
   } catch (error) {
-    logger.error(`Error clearing model cache for project ${projectId}`, error);
+    logger.error(
+      `Error clearing model cache for project ${projectId}: ${error}`,
+    );
+  }
+}
+
+export async function isModelMatchCacheLocked() {
+  try {
+    return Boolean(await redis?.exists(MODEL_MATCH_CACHE_LOCKED_KEY));
+  } catch (err) {
+    logger.error("Failed to check whether model match is locked", err);
+
+    return false;
+  }
+}
+
+export async function clearFullModelCache() {
+  if (env.LANGFUSE_CACHE_MODEL_MATCH_ENABLED === "false" || !redis) {
+    return;
+  }
+
+  try {
+    // Use lock to protect for concurrent executions
+    // This function is called on worker startup, so we want to avoid all workers triggering this delete
+    if (await isModelMatchCacheLocked()) {
+      logger.info("Model cache clearing already in progress; skipping.");
+
+      return;
+    }
+
+    const startTime = Date.now();
+    logger.info("Clearing full model cache...");
+
+    const tenMinutesInSeconds = 60 * 10;
+    await redis.setex(
+      MODEL_MATCH_CACHE_LOCKED_KEY,
+      tenMinutesInSeconds,
+      "locked",
+    );
+
+    const pattern = getModelMatchKeyPrefix() + "*";
+
+    const keys =
+      env.REDIS_CLUSTER_ENABLED === "true"
+        ? (
+            await Promise.all(
+              (redis as Cluster)
+                .nodes("master")
+                .map((node) => node.keys(pattern) || []),
+            )
+          ).flat()
+        : await redis.keys(pattern);
+
+    if (keys.length > 0) {
+      await safeMultiDel(redis, keys);
+      logger.info(
+        `Cleared full model cache with ${keys.length} keys in ${Date.now() - startTime}ms.`,
+      );
+    } else {
+      logger.info(`No keys found for match pattern '${pattern}'`);
+    }
+  } catch (error) {
+    logger.error(`Error clearing full model cache: ${error}`);
+  } finally {
+    await redis?.del(MODEL_MATCH_CACHE_LOCKED_KEY);
   }
 }

packages/shared/src/server/llm/fetchLLMCompletion.ts

@@ -42,6 +42,7 @@ import {
 } from "./types";
 import { CallbackHandler } from "langfuse-langchain";
 import type { BaseCallbackHandler } from "@langchain/core/callbacks/base";
+import { HttpsProxyAgent } from "https-proxy-agent";
 
 const isLangfuseCloud = Boolean(env.NEXT_PUBLIC_LANGFUSE_CLOUD_REGION);
 
@@ -217,6 +218,10 @@ export async function fetchLLMCompletion(
     (m) => m.content.length > 0 || "tool_calls" in m,
   );
 
+  // Common proxy configuration for all adapters
+  const proxyUrl = env.HTTPS_PROXY;
+  const proxyAgent = proxyUrl ? new HttpsProxyAgent(proxyUrl) : undefined;
+
   let chatModel:
     | ChatOpenAI
     | ChatAnthropic
@@ -232,7 +237,11 @@ export async function fetchLLMCompletion(
       maxTokens: modelParams.max_tokens,
       topP: modelParams.top_p,
       callbacks: finalCallbacks,
-      clientOptions: { maxRetries, timeout: 1000 * 60 * 2 }, // 2 minutes timeout
+      clientOptions: {
+        maxRetries,
+        timeout: 1000 * 60 * 2, // 2 minutes timeout
+        ...(proxyAgent && { httpAgent: proxyAgent }),
+      },
     });
   } else if (modelParams.adapter === LLMAdapter.OpenAI) {
     chatModel = new ChatOpenAI({
@@ -247,6 +256,7 @@ export async function fetchLLMCompletion(
       configuration: {
         baseURL,
         defaultHeaders: extraHeaders,
+        ...(proxyAgent && { httpAgent: proxyAgent }),
       },
       timeout: 1000 * 60 * 2, // 2 minutes timeout
     });
@@ -264,6 +274,7 @@ export async function fetchLLMCompletion(
       timeout: 1000 * 60 * 2, // 2 minutes timeout
       configuration: {
         defaultHeaders: extraHeaders,
+        ...(proxyAgent && { httpAgent: proxyAgent }),
       },
     });
   } else if (modelParams.adapter === LLMAdapter.Bedrock) {
@@ -376,6 +387,7 @@ export async function fetchLLMCompletion(
           },
           configuration: {
             baseURL,
+            ...(proxyAgent && { httpAgent: proxyAgent }),
           },
           timeout: 1000 * 60 * 2, // 2 minutes timeout
         })

packages/shared/src/server/llm/testModelCall.ts

@@ -0,0 +1,52 @@
+import { z as zodV3 } from "zod/v3";
+import {
+  ChatMessageRole,
+  ChatMessageType,
+  LLMApiKeySchema,
+  type ModelConfig,
+} from "./types";
+import { decrypt } from "../../encryption";
+import { fetchLLMCompletion } from "./fetchLLMCompletion";
+import { decryptAndParseExtraHeaders } from "./utils";
+import z from "zod/v4";
+
+export const testModelCall = async ({
+  provider,
+  model,
+  apiKey,
+  prompt,
+  modelConfig,
+}: {
+  provider: string;
+  model: string;
+  apiKey: z.infer<typeof LLMApiKeySchema>;
+  prompt?: string;
+  modelConfig?: ModelConfig | null;
+}) => {
+  (
+    await fetchLLMCompletion({
+      streaming: false,
+      apiKey: decrypt(apiKey.secretKey), // decrypt the secret key
+      extraHeaders: decryptAndParseExtraHeaders(apiKey.extraHeaders),
+      baseURL: apiKey.baseURL ?? undefined,
+      messages: [
+        {
+          role: ChatMessageRole.User,
+          content: prompt ?? "mock content",
+          type: ChatMessageType.User,
+        },
+      ],
+      modelParams: {
+        provider: provider,
+        model: model,
+        adapter: apiKey.adapter,
+        ...modelConfig,
+      },
+      structuredOutputSchema: zodV3.object({
+        score: zodV3.string(),
+        reasoning: zodV3.string(),
+      }),
+      config: apiKey.config,
+    })
+  ).completion;
+};

packages/shared/src/server/llm/types.ts

@@ -294,6 +294,12 @@ export const openAIModels = [
   "gpt-4.1-mini-2025-04-14",
   "gpt-4.1-nano",
   "gpt-4.1-nano-2025-04-14",
+  "gpt-5",
+  "gpt-5-2025-08-07",
+  "gpt-5-mini",
+  "gpt-5-mini-2025-08-07",
+  "gpt-5-nano",
+  "gpt-5-nano-2025-08-07",
   "o3",
   "o3-2025-04-16",
   "o4-mini",

packages/shared/src/server/repositories/observations.ts

@@ -1508,6 +1508,15 @@ export const getGenerationsForPostHog = async function* (
   minTimestamp: Date,
   maxTimestamp: Date,
 ) {
+  // Determine which trace table to use based on experiment flag
+  const useAMT = env.LANGFUSE_EXPERIMENT_RETURN_NEW_RESULT === "true";
+  // Subtract 7d from minTimestamp to account for shift in query
+  const traceTable = useAMT
+    ? getTimeframesTracesAMT(
+        new Date(minTimestamp.getTime() - 7 * 24 * 60 * 60 * 1000),
+      )
+    : "traces";
+
   const query = `
     SELECT
       o.name as name,
@@ -1532,7 +1541,7 @@ export const getGenerationsForPostHog = async function* (
       t.tags as trace_tags,
       t.metadata['$posthog_session_id'] as posthog_session_id
     FROM observations o FINAL
-    LEFT JOIN traces t FINAL ON o.trace_id = t.id AND o.project_id = t.project_id
+    LEFT JOIN ${traceTable} t FINAL ON o.trace_id = t.id AND o.project_id = t.project_id
     WHERE o.project_id = {projectId: String}
     AND t.project_id = {projectId: String}
     AND o.start_time >= {minTimestamp: DateTime64(3)}
@@ -1554,6 +1563,7 @@ export const getGenerationsForPostHog = async function* (
       type: "observation",
       kind: "analytic",
       projectId,
+      experiment_amt: useAMT ? "new" : "original",
     },
     clickhouseConfigs: {
       request_timeout: env.LANGFUSE_CLICKHOUSE_DATA_EXPORT_REQUEST_TIMEOUT_MS,

packages/shared/src/server/repositories/scores.ts

@@ -33,6 +33,7 @@ import { ClickHouseClientConfigOptions } from "@clickhouse/client";
 import { recordDistribution } from "../instrumentation";
 import { prisma } from "../../db";
 import { measureAndReturn } from "../clickhouse/measureAndReturn";
+import { getTimeframesTracesAMT } from "./traces";
 
 export const searchExistingAnnotationScore = async (
   projectId: string,
@@ -1152,7 +1153,7 @@ export const getNumericScoreHistogram = async (
   const query = `
     select s.value
     from scores s
-    ${traceFilter ? `LEFT JOIN traces t ON s.trace_id = t.id AND t.project_id = s.project_id` : ""}
+    ${traceFilter ? `LEFT JOIN __TRACE_TABLE__ t ON s.trace_id = t.id AND t.project_id = s.project_id` : ""}
     WHERE s.project_id = {projectId: String}
     ${traceFilter ? `AND t.project_id = {projectId: String}` : ""}
     ${chFilterRes?.query ? `AND ${chFilterRes.query}` : ""}
@@ -1161,18 +1162,45 @@ export const getNumericScoreHistogram = async (
     ${limit !== undefined ? `limit {limit: Int32}` : ""}
   `;
 
-  return queryClickhouse<{ value: number }>({
-    query,
-    params: {
-      projectId,
-      limit,
-      ...(chFilterRes ? chFilterRes.params : {}),
+  // Extract timestamp from filter for AMT table selection
+  const timestampFilter = chFilter.find(
+    (f) => f.clickhouseTable === "traces" && f.field === "timestamp",
+  ) as TimeFilter | undefined;
+  const timestamp = timestampFilter?.value;
+
+  return measureAndReturn({
+    operationName: "getNumericScoreHistogram",
+    projectId,
+    minStartTime: timestamp,
+    input: {
+      params: {
+        projectId,
+        limit,
+        ...(chFilterRes ? chFilterRes.params : {}),
+      },
+      tags: {
+        feature: "tracing",
+        type: "score",
+        kind: "analytic",
+        projectId,
+        operation_name: "getNumericScoreHistogram",
+      },
+      timestamp,
     },
-    tags: {
-      feature: "tracing",
-      type: "score",
-      kind: "analytic",
-      projectId,
+    existingExecution: async (input) => {
+      return queryClickhouse<{ value: number }>({
+        query: query.replace("__TRACE_TABLE__", "traces"),
+        params: input.params,
+        tags: { ...input.tags, experiment_amt: "original" },
+      });
+    },
+    newExecution: async (input) => {
+      const traceAmt = getTimeframesTracesAMT(input.timestamp);
+      return queryClickhouse<{ value: number }>({
+        query: query.replace("__TRACE_TABLE__", traceAmt),
+        params: input.params,
+        tags: { ...input.tags, experiment_amt: "new" },
+      });
     },
   });
 };
@@ -1399,6 +1427,15 @@ export const getScoresForPostHog = async function* (
   minTimestamp: Date,
   maxTimestamp: Date,
 ) {
+  // Determine which trace table to use based on experiment flag
+  const useAMT = env.LANGFUSE_EXPERIMENT_RETURN_NEW_RESULT === "true";
+  // Subtract 7d from minTimestamp to account for shift in query
+  const traceTable = useAMT
+    ? getTimeframesTracesAMT(
+        new Date(minTimestamp.getTime() - 7 * 24 * 60 * 60 * 1000),
+      )
+    : "traces";
+
   const query = `    SELECT
       s.id as id,
       s.timestamp as timestamp,
@@ -1417,7 +1454,7 @@ export const getScoresForPostHog = async function* (
       s.metadata as metadata,
       t.metadata['$posthog_session_id'] as posthog_session_id
     FROM scores s FINAL
-    LEFT JOIN traces t FINAL ON s.trace_id = t.id AND s.project_id = t.project_id
+    LEFT JOIN ${traceTable} t FINAL ON s.trace_id = t.id AND s.project_id = t.project_id
     WHERE s.project_id = {projectId: String}
     AND t.project_id = {projectId: String}
     AND s.timestamp >= {minTimestamp: DateTime64(3)}
@@ -1438,6 +1475,7 @@ export const getScoresForPostHog = async function* (
       type: "score",
       kind: "analytic",
       projectId,
+      experiment_amt: useAMT ? "new" : "original",
     },
     clickhouseConfigs: {
       request_timeout: env.LANGFUSE_CLICKHOUSE_DATA_EXPORT_REQUEST_TIMEOUT_MS,

packages/shared/src/server/repositories/traces.ts

@@ -47,12 +47,16 @@ enum TracesAMTs {
  * for <= 29 days, we use traces_30d_amt,
  * for all other cases we use traces_all_amt.
  *
+ * If LANGFUSE_EXPERIMENT_WHITELISTED_AMT_TABLES is set, we only return timeframes
+ * that are whitelisted or fallback to the traces_all_amt.
+ *
  * @param fromTimestamp
  */
 export const getTimeframesTracesAMT = (
   fromTimestamp: Date | undefined,
 ): TracesAMTs => {
   if (!fromTimestamp) {
+    // The TracesAllAMT must always be returned if there is no timestamp.
     return TracesAMTs.TracesAllAMT;
   }
 
@@ -60,12 +64,21 @@ export const getTimeframesTracesAMT = (
   const diffInDays = Math.floor(
     (now.getTime() - fromTimestamp.getTime()) / (1000 * 60 * 60 * 24),
   );
+
+  let selectedTable: TracesAMTs;
   if (diffInDays <= 6) {
-    return TracesAMTs.Traces7dAMT;
+    selectedTable = TracesAMTs.Traces7dAMT;
   } else if (diffInDays <= 29) {
-    return TracesAMTs.Traces30dAMT;
+    selectedTable = TracesAMTs.Traces30dAMT;
+  } else {
+    selectedTable = TracesAMTs.TracesAllAMT;
   }
-  return TracesAMTs.TracesAllAMT;
+
+  // Check if the selected table is whitelisted, fallback to TracesAllAMT if not
+  return env.LANGFUSE_EXPERIMENT_WHITELISTED_AMT_TABLES.length === 0 ||
+    env.LANGFUSE_EXPERIMENT_WHITELISTED_AMT_TABLES.includes(selectedTable)
+    ? selectedTable
+    : TracesAMTs.TracesAllAMT;
 };
 
 /**
@@ -1736,6 +1749,10 @@ export const getTracesForBlobStorageExport = function (
   minTimestamp: Date,
   maxTimestamp: Date,
 ) {
+  // Determine which trace table to use based on experiment flag
+  const useAMT = env.LANGFUSE_EXPERIMENT_RETURN_NEW_RESULT === "true";
+  const traceTable = useAMT ? getTimeframesTracesAMT(minTimestamp) : "traces";
+
   const query = `
     SELECT
       id,
@@ -1748,12 +1765,12 @@ export const getTracesForBlobStorageExport = function (
       session_id,
       release,
       version,
-      public,
-      bookmarked,
+      ${useAMT ? "finalizeAggregation(public)" : "public"} as public,
+      ${useAMT ? "finalizeAggregation(bookmarked)" : "bookmarked"} as bookmarked,
       tags,
-      input,
-      output
-    FROM traces FINAL
+      ${useAMT ? "finalizeAggregation(input)" : "input"} as input,
+      ${useAMT ? "finalizeAggregation(output)" : "output"} as output
+    FROM ${traceTable} FINAL
     WHERE project_id = {projectId: String}
     AND timestamp >= {minTimestamp: DateTime64(3)}
     AND timestamp <= {maxTimestamp: DateTime64(3)}
@@ -1771,6 +1788,7 @@ export const getTracesForBlobStorageExport = function (
       type: "trace",
       kind: "analytic",
       projectId,
+      experiment_amt: useAMT ? "new" : "original",
     },
     clickhouseConfigs: {
       request_timeout: env.LANGFUSE_CLICKHOUSE_DATA_EXPORT_REQUEST_TIMEOUT_MS,
@@ -1783,6 +1801,10 @@ export const getTracesForPostHog = async function* (
   minTimestamp: Date,
   maxTimestamp: Date,
 ) {
+  // Determine which trace table to use based on experiment flag
+  const useAMT = env.LANGFUSE_EXPERIMENT_RETURN_NEW_RESULT === "true";
+  const traceTable = useAMT ? getTimeframesTracesAMT(minTimestamp) : "traces";
+
   const query = `
     WITH observations_agg AS (
       SELECT o.project_id,
@@ -1810,7 +1832,7 @@ export const getTracesForPostHog = async function* (
       o.total_cost as total_cost,
       o.latency_milliseconds / 1000 as latency,
       o.observation_count as observation_count
-    FROM traces t FINAL
+    FROM ${traceTable} t FINAL
     LEFT JOIN observations_agg o ON t.id = o.trace_id AND t.project_id = o.project_id
     WHERE t.project_id = {projectId: String}
     AND t.timestamp >= {minTimestamp: DateTime64(3)}
@@ -1829,6 +1851,7 @@ export const getTracesForPostHog = async function* (
       type: "trace",
       kind: "analytic",
       projectId,
+      experiment_amt: useAMT ? "new" : "original",
     },
     clickhouseConfigs: {
       request_timeout: env.LANGFUSE_CLICKHOUSE_DATA_EXPORT_REQUEST_TIMEOUT_MS,

packages/shared/src/server/services/DefaultEvaluationModelService/DefaultEvalModelService.ts

@@ -1,7 +1,12 @@
 import z from "zod/v4";
 import { prisma } from "../../../db";
-import { LangfuseNotFoundError, QUEUE_ERROR_MESSAGES } from "../../../errors";
+import {
+  ForbiddenError,
+  LangfuseNotFoundError,
+  QUEUE_ERROR_MESSAGES,
+} from "../../../errors";
 import { LLMApiKeySchema, ZodModelConfig } from "../../llm/types";
+import { testModelCall } from "../../llm/testModelCall";
 
 type ValidConfig = {
   provider: string;
@@ -47,6 +52,23 @@ export class DefaultEvalModelService {
       );
     }
 
+    try {
+      if (LLMApiKeySchema.safeParse(llmApiKey).success) {
+        // Make a test structured output call to validate the LLM key
+        await testModelCall({
+          provider,
+          model,
+          apiKey: llmApiKey as z.infer<typeof LLMApiKeySchema>,
+          modelConfig: modelParams,
+        });
+      }
+    } catch (err) {
+      const message = err instanceof Error ? err.message : "Unknown error";
+      throw new ForbiddenError(
+        `Model configuration not valid for evaluation. ${message}`,
+      );
+    }
+
     // Create or update the default model
     return prisma.defaultLlmModel.upsert({
       where: {

packages/shared/src/server/test-utils/org-factory.ts

@@ -15,6 +15,7 @@ export function createBasicAuthHeader(
 
 export type CreateOrgProjectAndApiKeyOptions = {
   projectId?: string;
+  plan?: "Team" | "Hobby" | "Core" | "Pro" | "Enterprise";
 };
 export const createOrgProjectAndApiKey = async (
   props?: CreateOrgProjectAndApiKeyOptions,
@@ -25,7 +26,7 @@ export const createOrgProjectAndApiKey = async (
       id: v4(),
       name: v4(),
       cloudConfig: CloudConfigSchema.parse({
-        plan: "Team",
+        plan: props?.plan ?? "Team",
       }),
     },
   });

web/package.json

@@ -1,6 +1,6 @@
 {
   "name": "web",
-  "version": "3.96.2",
+  "version": "3.97.4",
   "private": true,
   "license": "MIT",
   "engines": {

web/public/generated/api/openapi.yml

@@ -79,6 +79,52 @@ paths:
               schema: {}
       security:
         - BasicAuth: []
+    post:
+      description: Create an annotation queue
+      operationId: annotationQueues_createQueue
+      tags:
+        - AnnotationQueues
+      parameters: []
+      responses:
+        '200':
+          description: ''
+          content:
+            application/json:
+              schema:
+                $ref: '#/components/schemas/AnnotationQueue'
+        '400':
+          description: ''
+          content:
+            application/json:
+              schema: {}
+        '401':
+          description: ''
+          content:
+            application/json:
+              schema: {}
+        '403':
+          description: ''
+          content:
+            application/json:
+              schema: {}
+        '404':
+          description: ''
+          content:
+            application/json:
+              schema: {}
+        '405':
+          description: ''
+          content:
+            application/json:
+              schema: {}
+      security:
+        - BasicAuth: []
+      requestBody:
+        required: true
+        content:
+          application/json:
+            schema:
+              $ref: '#/components/schemas/CreateAnnotationQueueRequest'
   /api/public/annotation-queues/{queueId}:
     get:
       description: Get an annotation queue by ID
@@ -3254,7 +3300,7 @@ paths:
       parameters:
         - name: filter
           in: query
-          description: Filter expression (e.g. userName eq 'value')
+          description: Filter expression (e.g. userName eq "value")
           required: false
           schema:
             type: string
@@ -4459,6 +4505,22 @@ components:
       required:
         - data
         - meta
+    CreateAnnotationQueueRequest:
+      title: CreateAnnotationQueueRequest
+      type: object
+      properties:
+        name:
+          type: string
+        description:
+          type: string
+          nullable: true
+        scoreConfigIds:
+          type: array
+          items:
+            type: string
+      required:
+        - name
+        - scoreConfigIds
     CreateAnnotationQueueItemRequest:
       title: CreateAnnotationQueueItemRequest
       type: object

web/public/generated/postman/collection.json

@@ -78,6 +78,39 @@
           },
           "response": []
         },
+        {
+          "_type": "endpoint",
+          "name": "Create Queue",
+          "request": {
+            "description": "Create an annotation queue",
+            "url": {
+              "raw": "{{baseUrl}}/api/public/annotation-queues",
+              "host": [
+                "{{baseUrl}}"
+              ],
+              "path": [
+                "api",
+                "public",
+                "annotation-queues"
+              ],
+              "query": [],
+              "variable": []
+            },
+            "header": [],
+            "method": "POST",
+            "auth": null,
+            "body": {
+              "mode": "raw",
+              "raw": "{\n    \"name\": \"example\",\n    \"description\": \"example\",\n    \"scoreConfigIds\": [\n        \"example\"\n    ]\n}",
+              "options": {
+                "raw": {
+                  "language": "json"
+                }
+              }
+            }
+          },
+          "response": []
+        },
         {
           "_type": "endpoint",
           "name": "Get Queue",

web/src/__tests__/annotation-queues-api.servertest.ts

@@ -14,6 +14,7 @@ import {
   CreateAnnotationQueueItemResponse,
   UpdateAnnotationQueueItemResponse,
   DeleteAnnotationQueueItemResponse,
+  CreateAnnotationQueueResponse,
 } from "@/src/features/public-api/types/annotation-queues";
 import {
   AnnotationQueueObjectType,
@@ -173,6 +174,120 @@ describe("Annotation Queues API Endpoints", () => {
     });
   });
 
+  describe("POST /annotation-queues", () => {
+    it("should create a new annotation queue", async () => {
+      const scoreConfig = await prisma.scoreConfig.create({
+        data: {
+          name: "Test Score Config",
+          description: "Test Score Config Description",
+          projectId,
+          dataType: "NUMERIC",
+        },
+      });
+
+      const response = await makeZodVerifiedAPICall(
+        CreateAnnotationQueueResponse,
+        "POST",
+        "/api/public/annotation-queues",
+        {
+          name: "Test Queue",
+          description: "Test Queue Description",
+          scoreConfigIds: [scoreConfig.id],
+        },
+        auth,
+      );
+
+      expect(response.status).toBe(200);
+      expect(response.body.id).toBeDefined();
+      expect(response.body.name).toBe("Test Queue");
+      expect(response.body.description).toBe("Test Queue Description");
+      expect(response.body.scoreConfigIds).toEqual([scoreConfig.id]);
+    });
+
+    it("should return 400 if the queue name already exists", async () => {
+      const response = await makeAPICall(
+        "POST",
+        "/api/public/annotation-queues",
+        {
+          name: "Test Queue",
+          description: "Test Queue Description",
+          scoreConfigIds: [],
+        },
+        auth,
+      );
+
+      expect(response.status).toBe(400);
+    });
+
+    it("should return 400 if no score config IDs are provided", async () => {
+      const response = await makeAPICall(
+        "POST",
+        "/api/public/annotation-queues",
+        {
+          name: "No configs queue",
+          description: "Test Queue Description",
+          scoreConfigIds: [],
+        },
+        auth,
+      );
+
+      expect(response.status).toBe(400);
+    });
+
+    it("should return 400 if the score config IDs are invalid", async () => {
+      const response = await makeAPICall(
+        "POST",
+        "/api/public/annotation-queues",
+        {
+          name: "Invalid configs queue",
+          description: "Test Queue Description",
+          scoreConfigIds: ["invalid-score-config-id"],
+        },
+        auth,
+      );
+
+      expect(response.status).toBe(400);
+    });
+
+    it("should return 405 if the user is on the Hobby plan and has reached the maximum number of annotation queues", async () => {
+      const { auth: hobbyPlanAuth, projectId: hobbyProjectId } =
+        await createOrgProjectAndApiKey({
+          plan: "Hobby",
+        });
+
+      const config = await prisma.scoreConfig.create({
+        data: {
+          name: "Test Score Config",
+          description: "Test Score Config Description",
+          projectId: hobbyProjectId,
+          dataType: "NUMERIC",
+        },
+      });
+
+      await prisma.annotationQueue.create({
+        data: {
+          name: "First queue",
+          description: "First queue description",
+          scoreConfigIds: [config.id],
+          projectId: hobbyProjectId,
+        },
+      });
+
+      const response = await makeAPICall(
+        "POST",
+        "/api/public/annotation-queues",
+        {
+          name: "Hobby plan queue",
+          description: "Test Queue Description",
+          scoreConfigIds: [config.id],
+        },
+        hobbyPlanAuth,
+      );
+
+      expect(response.status).toBe(405);
+    });
+  });
+
   describe("GET /annotation-queues/:queueId", () => {
     it("should get a specific annotation queue", async () => {
       const response = await makeZodVerifiedAPICall(

web/src/__tests__/async/metrics-api.servertest.ts

@@ -497,4 +497,137 @@ describe("/api/public/metrics API Endpoint", () => {
       expect(height).toBeGreaterThanOrEqual(0);
     });
   });
+
+  describe("LFE-6148: Comprehensive filter validation", () => {
+    it("should return 400 error for invalid array field filters", async () => {
+      // Test using string type on array field (tags) - should return validation error
+      const invalidStringTypeQuery = {
+        view: "traces",
+        dimensions: [{ field: "name" }],
+        metrics: [{ measure: "count", aggregation: "count" }],
+        filters: [
+          {
+            column: "tags",
+            operator: "contains",
+            value: "test-tag",
+            type: "string", // Invalid: array fields require arrayOptions type
+          },
+        ],
+        timeDimension: {
+          granularity: "day",
+        },
+        fromTimestamp: yesterday.toISOString(),
+        toTimestamp: tomorrow.toISOString(),
+        orderBy: null,
+      };
+
+      // Make API call and expect 400 error
+      const response = await makeAPICall(
+        "GET",
+        `/api/public/metrics?query=${encodeURIComponent(JSON.stringify(invalidStringTypeQuery))}`,
+      );
+
+      expect(response.status).toBe(400);
+      expect(response.body).toMatchObject({
+        error: "InvalidRequestError",
+        message: expect.stringContaining(
+          "Array fields require type 'arrayOptions', not 'string'",
+        ),
+      });
+    });
+
+    it("should return 400 error for invalid metadata filters", async () => {
+      // Test using wrong type for metadata field
+      const invalidMetadataTypeQuery = {
+        view: "traces",
+        dimensions: [{ field: "name" }],
+        metrics: [{ measure: "count", aggregation: "count" }],
+        filters: [
+          {
+            column: "metadata",
+            operator: "contains",
+            value: "test-value",
+            type: "string", // Invalid: metadata requires stringObject type
+          },
+        ],
+        timeDimension: null,
+        fromTimestamp: yesterday.toISOString(),
+        toTimestamp: tomorrow.toISOString(),
+        orderBy: null,
+      };
+
+      const response = await makeAPICall(
+        "GET",
+        `/api/public/metrics?query=${encodeURIComponent(JSON.stringify(invalidMetadataTypeQuery))}`,
+      );
+
+      expect(response.status).toBe(400);
+      expect(response.body).toMatchObject({
+        error: "InvalidRequestError",
+        message: expect.stringContaining(
+          "Metadata filters require type 'stringObject'",
+        ),
+      });
+    });
+
+    it("should work correctly with proper array field filter configuration", async () => {
+      // Setup test data with tags
+      const taggedTraceId = randomUUID();
+      await createTracesCh([
+        createTrace({
+          id: taggedTraceId,
+          name: "tagged-trace",
+          project_id: projectId,
+          timestamp: now.getTime(),
+          tags: ["test-tag", "another-tag"],
+          metadata: { test: testMetadataValue },
+        }),
+      ]);
+
+      // Test with correct arrayOptions filter
+      const validQuery = {
+        view: "traces",
+        dimensions: [{ field: "name" }],
+        metrics: [{ measure: "count", aggregation: "count" }],
+        filters: [
+          {
+            column: "tags",
+            operator: "any of", // Correct operator for array fields
+            value: ["test-tag"],
+            type: "arrayOptions", // Correct type for array fields
+          },
+          {
+            column: "metadata",
+            operator: "contains",
+            key: "test",
+            value: testMetadataValue,
+            type: "stringObject",
+          },
+        ],
+        timeDimension: null,
+        fromTimestamp: yesterday.toISOString(),
+        toTimestamp: tomorrow.toISOString(),
+        orderBy: null,
+      };
+
+      // Make the API call
+      const response = await makeZodVerifiedAPICall(
+        GetMetricsV1Response,
+        "GET",
+        `/api/public/metrics?query=${encodeURIComponent(JSON.stringify(validQuery))}`,
+      );
+
+      // Should succeed and return data
+      expect(response.status).toBe(200);
+      expect(Array.isArray(response.body.data)).toBe(true);
+      expect(response.body.data.length).toBeGreaterThan(0);
+
+      // Verify we got the tagged trace
+      const taggedTraceResult = response.body.data.find(
+        (row: any) => row.name === "tagged-trace",
+      );
+      expect(taggedTraceResult).toBeDefined();
+      expect(taggedTraceResult.count_count).toBe("1");
+    });
+  });
 });

web/src/constants/VERSION.ts

@@ -1 +1 @@
-export const VERSION = "v3.96.2";
+export const VERSION = "v3.97.4";

web/src/features/annotation-queues/components/UserAssignmentSection.tsx

@@ -5,6 +5,7 @@ import { Button } from "@/src/components/ui/button";
 import { MultiSelectCombobox } from "@/src/components/ui/multi-select-combobox";
 import { useUserSearch } from "@/src/features/annotation-queues/hooks/useUserSearch";
 import { useSelectedUsers } from "@/src/features/annotation-queues/hooks/useSelectedUsers";
+import { showSuccessToast } from "@/src/features/notifications/showSuccessToast";
 
 interface UserAssignmentSectionProps {
   projectId: string;
@@ -27,6 +28,7 @@ export const UserAssignmentSection = ({
     projectId: projectId,
     scope: "annotationQueueAssignments:CUD",
   });
+  const utils = api.useUtils();
 
   // Get current assigned users
   const queueAssignmentsQuery =
@@ -35,6 +37,18 @@ export const UserAssignmentSection = ({
       { enabled: !!queueId && hasQueueAssignmentsReadAccess },
     );
 
+  const deleteQueueAssignmentMutation =
+    api.annotationQueueAssignments.delete.useMutation({
+      onSuccess: () => {
+        utils.annotationQueueAssignments.invalidate();
+        utils.annotationQueues.invalidate();
+        showSuccessToast({
+          title: "Removed assignment",
+          description: "User removed from queue successfully",
+        });
+      },
+    });
+
   // Combine selected users and assigned users for exclusion
   const assignedUserIds =
     queueAssignmentsQuery.data?.assignments.map((user: any) => user.id) || [];
@@ -56,6 +70,16 @@ export const UserAssignmentSection = ({
     onChange(userIds);
   };
 
+  // Handle user removal
+  const handleUserRemove = (userId: string) => {
+    if (!!queueId)
+      deleteQueueAssignmentMutation.mutate({
+        projectId,
+        queueId,
+        userId,
+      });
+  };
+
   // Check if there are more assigned users than shown
   const hasMoreAssignedUsers =
     queueAssignmentsQuery.data &&
@@ -133,9 +157,17 @@ export const UserAssignmentSection = ({
                           </p>
                         </div>
                       </div>
-                      {/* <Button variant="ghost" size="icon-sm">
+                      <Button
+                        variant="ghost"
+                        size="icon-sm"
+                        disabled={
+                          !hasQueueAssignmentWriteAccess ||
+                          deleteQueueAssignmentMutation.isLoading
+                        }
+                        onClick={() => handleUserRemove(user.id)}
+                      >
                         <X className="h-3 w-3" />
-                      </Button> */}
+                      </Button>
                     </div>
                     {(index <
                       queueAssignmentsQuery.data?.assignments.length - 1 ||

web/src/features/datasets/components/DatasetActionButton.tsx

@@ -7,7 +7,7 @@ import {
   DialogHeader,
   DialogTitle,
 } from "@/src/components/ui/dialog";
-import { useState } from "react";
+import { useState, forwardRef } from "react";
 import { DialogTrigger } from "@radix-ui/react-dialog";
 import { DatasetForm } from "@/src/features/datasets/components/DatasetForm";
 import { useHasProjectAccess } from "@/src/features/rbac/utils/checkProjectAccess";
@@ -46,7 +46,10 @@ type DatasetActionButtonProps =
   | UpdateDatasetButtonProps
   | DeleteDatasetButtonProps;
 
-export const DatasetActionButton = (props: DatasetActionButtonProps) => {
+export const DatasetActionButton = forwardRef<
+  HTMLButtonElement,
+  DatasetActionButtonProps
+>((props, ref) => {
   const capture = usePostHogClientCapture();
   const [open, setOpen] = useState(false);
   const hasAccess = useHasProjectAccess({
@@ -60,6 +63,7 @@ export const DatasetActionButton = (props: DatasetActionButtonProps) => {
         {props.mode === "update" ? (
           props.icon ? (
             <Button
+              ref={ref}
               variant={props.variant || "outline"}
               size={props.size || "icon"}
               className={props.className}
@@ -74,6 +78,7 @@ export const DatasetActionButton = (props: DatasetActionButtonProps) => {
             </Button>
           ) : (
             <Button
+              ref={ref}
               variant={props.variant || "ghost"}
               size={props.size || "icon"}
               className={props.className}
@@ -95,6 +100,7 @@ export const DatasetActionButton = (props: DatasetActionButtonProps) => {
           )
         ) : props.mode === "delete" ? (
           <Button
+            ref={ref}
             variant={props.variant || "ghost"}
             size={props.size}
             className={props.className}
@@ -115,6 +121,7 @@ export const DatasetActionButton = (props: DatasetActionButtonProps) => {
           </Button>
         ) : (
           <Button
+            ref={ref}
             size={props.size}
             className={props.className}
             disabled={!hasAccess}
@@ -175,4 +182,6 @@ export const DatasetActionButton = (props: DatasetActionButtonProps) => {
       </DialogContent>
     </Dialog>
   );
-};
+});
+
+DatasetActionButton.displayName = "DatasetActionButton";

web/src/features/datasets/components/DatasetItemsTable.tsx

@@ -2,6 +2,7 @@ import { DataTable } from "@/src/components/table/data-table";
 import TableLink from "@/src/components/table/table-link";
 import { api } from "@/src/utils/api";
 import { type RouterOutput } from "@/src/utils/types";
+import { useRouter } from "next/router";
 import {
   DropdownMenu,
   DropdownMenuContent,
@@ -10,7 +11,7 @@ import {
   DropdownMenuTrigger,
 } from "@/src/components/ui/dropdown-menu";
 import { useQueryParams, withDefault, NumberParam } from "use-query-params";
-import { Archive, ListTree, MoreVertical, Trash2 } from "lucide-react";
+import { Archive, Edit, ListTree, MoreVertical, Trash2 } from "lucide-react";
 import { Button } from "@/src/components/ui/button";
 import { type DatasetItem, DatasetStatus, type Prisma } from "@langfuse/shared";
 import { type LangfuseColumnDef } from "@/src/components/table/types";
@@ -53,6 +54,7 @@ export function DatasetItemsTable({
   datasetId: string;
   menuItems?: React.ReactNode;
 }) {
+  const router = useRouter();
   const { setDetailPageList } = useDetailPageLists();
   const utils = api.useUtils();
   const capture = usePostHogClientCapture();
@@ -229,6 +231,17 @@ export function DatasetItemsTable({
             </DropdownMenuTrigger>
             <DropdownMenuContent align="end">
               <DropdownMenuLabel>Actions</DropdownMenuLabel>
+              <DropdownMenuItem
+                disabled={!hasAccess}
+                onClick={() => {
+                  router.push(
+                    `/project/${projectId}/datasets/${datasetId}/items/${id}`,
+                  );
+                }}
+              >
+                <Edit className="mr-2 h-4 w-4" />
+                Edit
+              </DropdownMenuItem>
               <DropdownMenuItem
                 disabled={!hasAccess}
                 onClick={() => {

web/src/features/datasets/lib/csvHelpers.ts

@@ -214,6 +214,15 @@ export function parseColumns(
   headerMap: Map<string, number>,
 ): Prisma.JsonValue {
   if (columnNames.length === 0) return null;
+
+  // Single column: do not nest columns into json objects
+  if (columnNames.length === 1) {
+    const col = columnNames[0];
+    const rawValue = row[headerMap.get(col)!];
+    return parseValue(rawValue);
+  }
+
+  // Multiple columns: nest columns into json objects
   return Object.fromEntries(
     columnNames.map((col) => [col, parseValue(row[headerMap.get(col)!])]),
   );

web/src/features/evals/components/eval-template-detail.tsx

@@ -33,9 +33,6 @@ export const EvalTemplateDetail = () => {
   const templateId = router.query.id as string;
 
   const [isEditing, setIsEditing] = useState(false);
-  const [selectedTemplate, setSelectedTemplate] = useState<EvalTemplate | null>(
-    null,
-  );
 
   // get the current template by id
   const template = api.evals.templateById.useQuery({
@@ -58,15 +55,7 @@ export const EvalTemplateDetail = () => {
     },
   );
 
-  // Set the selected template when data is loaded
-  React.useEffect(() => {
-    if (template.data && !selectedTemplate) {
-      setSelectedTemplate(template.data);
-    }
-  }, [template.data, selectedTemplate]);
-
   const handleTemplateSelect = (newTemplate: EvalTemplate) => {
-    setSelectedTemplate(newTemplate);
     // Update URL without full page reload
     router.push(
       `/project/${projectId}/evals/templates/${newTemplate.id}`,
@@ -75,13 +64,10 @@ export const EvalTemplateDetail = () => {
     );
   };
 
-  // Get the appropriate template to display
-  const displayTemplate = selectedTemplate || template.data;
-
   return (
     <Page
       headerProps={{
-        title: `${displayTemplate?.name || ""}`,
+        title: `${template.data?.name ?? ""}`,
         itemType: "EVALUATOR",
         breadcrumb: [
           {
@@ -95,7 +81,7 @@ export const EvalTemplateDetail = () => {
               projectId={projectId}
               isEditing={isEditing}
               setIsEditing={setIsEditing}
-              isCustom={!!displayTemplate?.projectId}
+              isCustom={!!template.data?.projectId}
             />
 
             {/* TODO: moved to LFE-4573 */}
@@ -114,14 +100,14 @@ export const EvalTemplateDetail = () => {
         ),
       }}
     >
-      {allTemplates.isLoading || !allTemplates.data || !displayTemplate ? (
+      {allTemplates.isLoading || !allTemplates.data || !template.data ? (
         <div className="p-3">Loading...</div>
       ) : isEditing ? (
         <div className="overflow-y-auto p-3 pt-1">
           <EvalTemplateForm
             useDialog={false}
             projectId={projectId}
-            existingEvalTemplate={displayTemplate}
+            existingEvalTemplate={template.data}
             isEditing={isEditing}
             setIsEditing={setIsEditing}
           />
@@ -132,7 +118,7 @@ export const EvalTemplateDetail = () => {
             <EvalTemplateForm
               useDialog={false}
               projectId={projectId}
-              existingEvalTemplate={displayTemplate}
+              existingEvalTemplate={template.data}
               isEditing={isEditing}
               setIsEditing={setIsEditing}
             />
@@ -150,7 +136,7 @@ export const EvalTemplateDetail = () => {
                     <div
                       key={template.id}
                       className={`flex cursor-pointer flex-col rounded-md px-2 py-1.5 hover:bg-accent ${
-                        template.id === displayTemplate.id ? "bg-accent" : ""
+                        template.id === templateId ? "bg-accent" : ""
                       }`}
                       onClick={() => handleTemplateSelect(template)}
                     >

web/src/features/evals/components/evaluator-selector.tsx

@@ -58,6 +58,12 @@ export function EvaluatorSelector({
     },
   );
 
+  // Ensure per-name arrays are sorted by createdAt ascending so last is latest
+  const sortByCreatedAt = (arr: EvalTemplate[]) =>
+    arr.sort((a, b) => a.createdAt.getTime() - b.createdAt.getTime());
+  Object.values(groupedTemplates.custom).forEach(sortByCreatedAt);
+  Object.values(groupedTemplates.langfuse).forEach(sortByCreatedAt);
+
   // Filter templates based on search
   const filteredTemplates = {
     langfuse: Object.entries(groupedTemplates.langfuse)

web/src/features/evals/pages/default-evaluation-model.tsx

@@ -30,6 +30,7 @@ export default function DefaultEvaluationModelPage() {
   const projectId = router.query.projectId as string;
   const utils = api.useUtils();
   const [isEditing, setIsEditing] = useState(false);
+  const [formError, setFormError] = useState<string | null>(null);
 
   const hasWriteAccess = useHasProjectAccess({
     projectId,
@@ -56,7 +57,7 @@ export default function DefaultEvaluationModelPage() {
     setModelParams,
   );
 
-  const { mutate: upsertDefaultModel, isLoading } =
+  const { mutateAsync: upsertDefaultModel, isLoading } =
     api.defaultLlmModel.upsertDefaultModel.useMutation({
       onSuccess: () => {
         showSuccessToast({
@@ -65,26 +66,26 @@ export default function DefaultEvaluationModelPage() {
         });
 
         utils.defaultLlmModel.fetchDefaultModel.invalidate({ projectId });
+        setFormError(null);
+        setIsEditing(false);
+      },
+      onError: (error) => {
+        setFormError(error.message as string);
       },
     });
 
-  const executeUpsertMutation = () => {
-    try {
-      upsertDefaultModel({
-        projectId,
-        provider: modelParams.provider.value,
-        adapter: modelParams.adapter.value,
-        model: modelParams.model.value,
-        modelParams: {
-          max_tokens: modelParams.max_tokens.value,
-          temperature: modelParams.temperature.value,
-          top_p: modelParams.top_p.value,
-        },
-      });
-    } catch (error) {
-      return Promise.reject(error);
-    }
-    setIsEditing(false);
+  const executeUpsertMutation = async () => {
+    await upsertDefaultModel({
+      projectId,
+      provider: modelParams.provider.value,
+      adapter: modelParams.adapter.value,
+      model: modelParams.model.value,
+      modelParams: {
+        max_tokens: modelParams.max_tokens.value,
+        temperature: modelParams.temperature.value,
+        top_p: modelParams.top_p.value,
+      },
+    });
   };
 
   if (isDefaultModelLoading) {
@@ -135,7 +136,15 @@ export default function DefaultEvaluationModelPage() {
             />
           )}
 
-          <Dialog open={isEditing} onOpenChange={setIsEditing}>
+          <Dialog
+            open={isEditing}
+            onOpenChange={(open) => {
+              setIsEditing(open);
+              if (!open) {
+                setFormError(null);
+              }
+            }}
+          >
             <DialogTrigger asChild>
               <Button
                 disabled={!hasWriteAccess}
@@ -167,24 +176,31 @@ export default function DefaultEvaluationModelPage() {
               <div className="my-2 text-xs text-muted-foreground">
                 Select a model which supports function calling.
               </div>
-              <div className="mt-2 flex justify-end gap-2">
-                <Button variant="outline" onClick={() => setIsEditing(false)}>
-                  Cancel
-                </Button>
-                {selectedModel ? (
-                  <UpdateButton
-                    projectId={projectId}
-                    isLoading={isLoading}
-                    executeUpsertMutation={executeUpsertMutation}
-                  />
-                ) : (
-                  <Button
-                    disabled={!hasWriteAccess || !modelParams.provider.value}
-                    onClick={executeUpsertMutation}
-                  >
-                    Save
+              <div className="flex flex-col gap-2">
+                <div className="mt-2 flex justify-end gap-2">
+                  <Button variant="outline" onClick={() => setIsEditing(false)}>
+                    Cancel
                   </Button>
-                )}
+                  {selectedModel ? (
+                    <UpdateButton
+                      projectId={projectId}
+                      isLoading={isLoading}
+                      executeUpsertMutation={executeUpsertMutation}
+                    />
+                  ) : (
+                    <Button
+                      disabled={!hasWriteAccess || !modelParams.provider.value}
+                      onClick={executeUpsertMutation}
+                    >
+                      Save
+                    </Button>
+                  )}
+                </div>
+                {formError ? (
+                  <p className="text-red w-full text-center">
+                    <span className="font-bold">Error:</span> {formError}
+                  </p>
+                ) : null}
               </div>
             </DialogContent>
           </Dialog>
@@ -223,7 +239,10 @@ function UpdateButton({
           Update
         </Button>
       </PopoverTrigger>
-      <PopoverContent onClick={(e) => e.stopPropagation()}>
+      <PopoverContent
+        onClick={(e) => e.stopPropagation()}
+        className="w-fit max-w-[500px]"
+      >
         <h2 className="text-md mb-3 font-semibold">Please confirm</h2>
         <p className="mb-3 text-sm">
           Updating the default model will impact any currently running

web/src/features/evals/server/defaultEvalModelRouter.ts

@@ -5,6 +5,7 @@ import {
 } from "@/src/server/api/trpc";
 import { z } from "zod/v4";
 import {
+  ForbiddenError,
   InvalidRequestError,
   LangfuseNotFoundError,
   ZodModelConfig,
@@ -42,12 +43,14 @@ export const defaultEvalModelRouter = createTRPCRouter({
       });
 
       try {
-        return DefaultEvalModelService.upsertDefaultModel(input);
+        return await DefaultEvalModelService.upsertDefaultModel(input);
       } catch (error) {
         if (error instanceof InvalidRequestError) {
           throw new TRPCError({ code: "BAD_REQUEST", message: error.message });
         } else if (error instanceof LangfuseNotFoundError) {
           throw new TRPCError({ code: "NOT_FOUND", message: error.message });
+        } else if (error instanceof ForbiddenError) {
+          throw new TRPCError({ code: "FORBIDDEN", message: error.message });
         }
         throw error;
       }

web/src/features/evals/server/router.ts

@@ -1,5 +1,4 @@
 import { z } from "zod/v4";
-import { z as zodV3 } from "zod/v3";
 import {
   createTRPCRouter,
   protectedProjectProcedure,
@@ -11,7 +10,6 @@ import {
   ZodModelConfig,
   singleFilter,
   variableMapping,
-  ChatMessageRole,
   paginationZod,
   type JobConfiguration,
   JobType,
@@ -21,19 +19,16 @@ import {
   orderBy,
   jsonSchema,
 } from "@langfuse/shared";
-import { decrypt } from "@langfuse/shared/encryption";
 import {
-  decryptAndParseExtraHeaders,
-  fetchLLMCompletion,
   getQueue,
   getScoresByIds,
   logger,
   QueueName,
   QueueJobs,
-  ChatMessageType,
   tableColumnsToSqlFilterAndPrefix,
   orderByToPrismaSql,
   DefaultEvalModelService,
+  testModelCall,
 } from "@langfuse/shared/src/server";
 import { TRPCError } from "@trpc/server";
 import { EvalReferencedEvaluators } from "@/src/features/evals/types";
@@ -808,45 +803,20 @@ export const evalRouter = createTRPCRouter({
         });
       }
 
-      const matchingLLMKey = modelConfig.config.apiKey;
-
-      // Make a test structured output call to validate the LLM key
       try {
-        (
-          await fetchLLMCompletion({
-            streaming: false,
-            apiKey: decrypt(matchingLLMKey.secretKey), // decrypt the secret key
-            extraHeaders: decryptAndParseExtraHeaders(
-              matchingLLMKey.extraHeaders,
-            ),
-            baseURL: matchingLLMKey.baseURL ?? undefined,
-            messages: [
-              {
-                role: ChatMessageRole.User,
-                content: input.prompt,
-                type: ChatMessageType.User,
-              },
-            ],
-            modelParams: {
-              provider: modelConfig.config.provider,
-              model: modelConfig.config.model,
-              adapter: matchingLLMKey.adapter,
-              ...input.modelParams,
-            },
-            structuredOutputSchema: zodV3.object({
-              score: zodV3.string(),
-              reasoning: zodV3.string(),
-            }),
-            config: matchingLLMKey.config,
-          })
-        ).completion;
+        // Make a test structured output call to validate the LLM key
+        await testModelCall({
+          provider: modelConfig.config.provider,
+          model: modelConfig.config.model,
+          apiKey: modelConfig.config.apiKey,
+          modelConfig: input.modelParams,
+          prompt: input.prompt,
+        });
       } catch (err) {
-        logger.error(err);
-
+        const message = err instanceof Error ? err.message : "Unknown error";
         throw new TRPCError({
           code: "PRECONDITION_FAILED",
-          message:
-            "Selected model is not supported for evaluations. Test tool call failed.",
+          message: `Model configuration not valid for evaluation. ${message}`,
         });
       }
 

web/src/features/llm-api-key/server/router.ts

@@ -289,6 +289,7 @@ export const llmApiKeyRouter = createTRPCRouter({
               customModels: true,
               withDefaultModels: true,
               extraHeaderKeys: true,
+              config: true,
             },
             where: {
               projectId: input.projectId,

web/src/features/organizations/components/NewOrganizationForm.tsx

@@ -22,7 +22,7 @@ import {
 } from "@/src/components/ui/select";
 import { api } from "@/src/utils/api";
 import { useSession } from "next-auth/react";
-import { organizationNameSchema } from "@/src/features/organizations/utils/organizationNameSchema";
+import { organizationFormSchema } from "@/src/features/organizations/utils/organizationNameSchema";
 import { usePostHogClientCapture } from "@/src/features/posthog-analytics/usePostHogClientCapture";
 import { SurveyName } from "@prisma/client";
 import { env } from "@/src/env.mjs";
@@ -35,7 +35,7 @@ export const NewOrganizationForm = ({
   const { update: updateSession } = useSession();
 
   const form = useForm({
-    resolver: zodResolver(organizationNameSchema),
+    resolver: zodResolver(organizationFormSchema),
     defaultValues: {
       name: "",
       type: "Personal",
@@ -50,7 +50,7 @@ export const NewOrganizationForm = ({
   const watchedType = form.watch("type");
   const isCloud = Boolean(env.NEXT_PUBLIC_LANGFUSE_CLOUD_REGION);
 
-  function onSubmit(values: z.infer<typeof organizationNameSchema>) {
+  function onSubmit(values: z.infer<typeof organizationFormSchema>) {
     capture("organizations:new_form_submit");
     createOrgMutation
       .mutateAsync({
@@ -130,13 +130,10 @@ export const NewOrganizationForm = ({
                   <FormDescription>
                     What would best describe your organization?
                   </FormDescription>
-                  <Select
-                    onValueChange={field.onChange}
-                    defaultValue={field.value}
-                  >
+                  <Select onValueChange={field.onChange} value={field.value}>
                     <FormControl>
-                      <SelectTrigger>
-                        <SelectValue placeholder="Select organization type" />
+                      <SelectTrigger ref={field.ref}>
+                        <SelectValue placeholder="Please choose" />
                       </SelectTrigger>
                     </FormControl>
                     <SelectContent>
@@ -162,13 +159,10 @@ export const NewOrganizationForm = ({
                     <FormDescription>
                       How many people are in your {watchedType}?
                     </FormDescription>
-                    <Select
-                      onValueChange={field.onChange}
-                      defaultValue={field.value}
-                    >
+                    <Select onValueChange={field.onChange} value={field.value}>
                       <FormControl>
-                        <SelectTrigger>
-                          <SelectValue placeholder="Select organization size" />
+                        <SelectTrigger ref={field.ref}>
+                          <SelectValue placeholder="Please choose" />
                         </SelectTrigger>
                       </FormControl>
                       <SelectContent>

web/src/features/organizations/utils/organizationNameSchema.ts

@@ -18,11 +18,29 @@ const organizationSizeOptions = [
   "More than 300",
 ] as const;
 
+// Base schema for org creation, used for server-side validation too
 export const organizationNameSchema = z.object({
   name: StringNoHTML.min(3, "Must have at least 3 characters").max(
     60,
     "Must have at most 60 characters",
   ),
-  type: z.enum(organizationTypeOptions).optional(),
-  size: z.enum(organizationSizeOptions).optional(),
 });
+
+// Extended schema for client-side form validation including type and size,
+// which are posted separately as a survey response.
+export const organizationFormSchema = organizationNameSchema
+  .extend({
+    type: z.enum(organizationTypeOptions),
+    size: z.enum(organizationSizeOptions).optional(),
+  })
+  .check((ctx) => {
+    const { type, size } = ctx.value;
+    if ((type === "Company" || type === "Agency") && !size) {
+      ctx.issues.push({
+        code: z.ZodIssueCode.custom,
+        path: ["size"],
+        input: ctx.value.size,
+        message: "Please specify the size of your organization",
+      });
+    }
+  });

web/src/features/public-api/components/CreateLLMApiKeyForm.tsx

@@ -40,6 +40,9 @@ import { env } from "@/src/env.mjs";
 
 const isLangfuseCloud = Boolean(env.NEXT_PUBLIC_LANGFUSE_CLOUD_REGION);
 
+const isCustomModelsRequired = (adapter: LLMAdapter) =>
+  adapter === LLMAdapter.Azure || adapter === LLMAdapter.Bedrock;
+
 const createFormSchema = (mode: "create" | "update") =>
   z
     .object({
@@ -62,11 +65,7 @@ const createFormSchema = (mode: "create" | "update") =>
         }),
       ),
     })
-    .refine((data) => data.withDefaultModels || data.customModels.length > 0, {
-      message:
-        "At least one custom model name is required when default models are disabled.",
-      path: ["withDefaultModels"],
-    })
+    // 1) If adapter requires custom models, enforce that first
     .refine(
       (data) => {
         if (data.adapter !== LLMAdapter.Bedrock) return true;
@@ -88,6 +87,32 @@ const createFormSchema = (mode: "create" | "update") =>
         path: ["adapter"],
       },
     )
+    .refine(
+      (data) => {
+        if (isCustomModelsRequired(data.adapter)) {
+          return data.customModels.length > 0;
+        }
+        return true;
+      },
+      {
+        message: "At least one custom model is required for this adapter.",
+        path: ["customModels"],
+      },
+    )
+    // 2) For adapters that support defaults, require default models or at least one custom model
+    .refine(
+      (data) => {
+        if (isCustomModelsRequired(data.adapter)) {
+          return true;
+        }
+        return data.withDefaultModels || data.customModels.length > 0;
+      },
+      {
+        message:
+          "At least one custom model name is required when default models are disabled.",
+        path: ["withDefaultModels"],
+      },
+    )
     .refine(
       (data) =>
         data.adapter === LLMAdapter.Bedrock ||
@@ -200,6 +225,12 @@ export function CreateLLMApiKeyForm({
 
   const currentAdapter = form.watch("adapter");
 
+  const hasAdvancedSettings = (adapter: LLMAdapter) =>
+    adapter === LLMAdapter.OpenAI ||
+    adapter === LLMAdapter.Anthropic ||
+    adapter === LLMAdapter.VertexAI ||
+    adapter === LLMAdapter.GoogleAIStudio;
+
   const { fields, append, remove } = useFieldArray({
     control: form.control,
     name: "customModels",
@@ -214,6 +245,114 @@ export function CreateLLMApiKeyForm({
     name: "extraHeaders",
   });
 
+  const renderCustomModelsField = () => (
+    <FormField
+      control={form.control}
+      name="customModels"
+      render={() => (
+        <FormItem>
+          <FormLabel>Custom models</FormLabel>
+          <FormDescription>
+            Custom model names accepted by given endpoint.
+          </FormDescription>
+          {currentAdapter === LLMAdapter.Azure && (
+            <FormDescription className="text-dark-yellow">
+              {
+                "For Azure, the model name should be the same as the deployment name in Azure. For evals, choose a model with function calling capabilities."
+              }
+            </FormDescription>
+          )}
+
+          {currentAdapter === LLMAdapter.Bedrock && (
+            <FormDescription className="text-dark-yellow">
+              {
+                "For Bedrock, the model name is the Bedrock Inference Profile ID, e.g. 'eu.anthropic.claude-3-5-sonnet-20240620-v1:0'"
+              }
+            </FormDescription>
+          )}
+
+          {fields.map((customModel, index) => (
+            <span key={customModel.id} className="flex flex-row space-x-2">
+              <Input
+                {...form.register(`customModels.${index}.value`)}
+                placeholder={`Custom model name ${index + 1}`}
+              />
+              <Button
+                type="button"
+                variant="ghost"
+                onClick={() => remove(index)}
+              >
+                <TrashIcon className="h-4 w-4" />
+              </Button>
+            </span>
+          ))}
+          <Button
+            type="button"
+            variant="ghost"
+            onClick={() => append({ value: "" })}
+            className="w-full"
+          >
+            <PlusIcon className="-ml-0.5 mr-1.5 h-5 w-5" aria-hidden="true" />
+            Add custom model name
+          </Button>
+        </FormItem>
+      )}
+    />
+  );
+
+  const renderExtraHeadersField = () => (
+    <FormField
+      control={form.control}
+      name="extraHeaders"
+      render={() => (
+        <FormItem>
+          <FormLabel>Extra Headers</FormLabel>
+          <FormDescription>
+            Optional additional HTTP headers to include with requests towards
+            LLM provider. All header values stored encrypted{" "}
+            {isLangfuseCloud ? "on our servers" : "in your database"}.
+          </FormDescription>
+
+          {headerFields.map((header, index) => (
+            <div key={header.id} className="flex flex-row space-x-2">
+              <Input
+                {...form.register(`extraHeaders.${index}.key`)}
+                placeholder="Header name"
+              />
+              <Input
+                {...form.register(`extraHeaders.${index}.value`)}
+                placeholder={
+                  mode === "update" &&
+                  existingKey?.extraHeaderKeys &&
+                  existingKey.extraHeaderKeys[index]
+                    ? "***"
+                    : "Header value"
+                }
+              />
+              <Button
+                type="button"
+                variant="ghost"
+                onClick={() => removeHeader(index)}
+              >
+                <TrashIcon className="h-4 w-4" />
+              </Button>
+            </div>
+          ))}
+
+          <Button
+            type="button"
+            variant="ghost"
+            onClick={() => appendHeader({ key: "", value: "" })}
+            className="w-full"
+          >
+            <PlusIcon className="-ml-0.5 mr-1.5 h-5 w-5" aria-hidden="true" />
+            Add Header
+          </Button>
+        </FormItem>
+      )}
+    />
+  );
+
   // Disable provider and adapter fields in update mode
   const isFieldDisabled = (fieldName: string) => {
     if (mode !== "update") return false;
@@ -285,14 +424,16 @@ export function CreateLLMApiKeyForm({
           )
         : undefined;
 
-    const newKey = {
+    const newLlmApiKey = {
       id: existingKey?.id ?? "",
       projectId,
       secretKey: secretKey ?? "",
       provider: values.provider,
       adapter: values.adapter,
       baseURL: values.baseURL || undefined,
-      withDefaultModels: values.withDefaultModels,
+      withDefaultModels: isCustomModelsRequired(currentAdapter)
+        ? false
+        : values.withDefaultModels,
       config,
       customModels: values.customModels
         .map((m) => m.value.trim())
@@ -303,8 +444,8 @@ export function CreateLLMApiKeyForm({
     try {
       const testResult =
         mode === "create"
-          ? await mutTestLLMApiKey.mutateAsync(newKey)
-          : await mutTestUpdateLLMApiKey.mutateAsync(newKey);
+          ? await mutTestLLMApiKey.mutateAsync(newLlmApiKey)
+          : await mutTestUpdateLLMApiKey.mutateAsync(newLlmApiKey);
 
       if (!testResult.success) throw new Error(testResult.error);
     } catch (error) {
@@ -320,7 +461,7 @@ export function CreateLLMApiKeyForm({
     }
 
     return (mode === "create" ? mutCreateLlmApiKey : mutUpdateLlmApiKey)
-      .mutateAsync(newKey)
+      .mutateAsync(newLlmApiKey)
       .then(() => {
         form.reset();
         onSuccess();
@@ -337,24 +478,6 @@ export function CreateLLMApiKeyForm({
         onSubmit={form.handleSubmit(onSubmit)}
       >
         <DialogBody>
-          {/* Provider name */}
-          <FormField
-            control={form.control}
-            name="provider"
-            render={({ field }) => (
-              <FormItem>
-                <FormLabel>Provider name</FormLabel>
-                <FormDescription>
-                  Name to identify the key within Langfuse.
-                </FormDescription>
-                <FormControl>
-                  <Input {...field} disabled={isFieldDisabled("provider")} />
-                </FormControl>
-                <FormMessage />
-              </FormItem>
-            )}
-          />
-
           {/* LLM adapter */}
           <FormField
             control={form.control}
@@ -393,6 +516,27 @@ export function CreateLLMApiKeyForm({
               </FormItem>
             )}
           />
+          {/* Provider name */}
+          <FormField
+            control={form.control}
+            name="provider"
+            render={({ field }) => (
+              <FormItem>
+                <FormLabel>Provider name</FormLabel>
+                <FormDescription>
+                  Key to identify the connection within Langfuse.
+                </FormDescription>
+                <FormControl>
+                  <Input
+                    {...field}
+                    placeholder={`e.g. ${currentAdapter}`}
+                    disabled={isFieldDisabled("provider")}
+                  />
+                </FormControl>
+                <FormMessage />
+              </FormItem>
+            )}
+          />
 
           {/* API Key or AWS Credentials */}
           {currentAdapter === LLMAdapter.Bedrock ? (
@@ -563,61 +707,64 @@ export function CreateLLMApiKeyForm({
             />
           )}
 
-          <div className="flex items-center">
-            <Button
-              type="button"
-              variant="link"
-              size="sm"
-              className="flex items-center pl-0"
-              onClick={() => setShowAdvancedSettings(!showAdvancedSettings)}
-            >
-              <span>
-                {showAdvancedSettings
-                  ? "Hide advanced settings"
-                  : "Show advanced settings"}
-              </span>
-              <ChevronDown
-                className={`ml-1 h-4 w-4 transition-transform ${showAdvancedSettings ? "rotate-180" : "rotate-0"}`}
-              />
-            </Button>
-          </div>
+          {/* Custom models: top-level for Azure/Bedrock */}
+          {isCustomModelsRequired(currentAdapter) && renderCustomModelsField()}
 
-          {showAdvancedSettings && (
+          {/* Extra headers - show for Azure in main section (Azure has no advanced settings) */}
+          {currentAdapter === LLMAdapter.Azure && renderExtraHeadersField()}
+
+          {hasAdvancedSettings(currentAdapter) && (
+            <div className="flex items-center">
+              <Button
+                type="button"
+                variant="link"
+                size="sm"
+                className="flex items-center pl-0"
+                onClick={() => setShowAdvancedSettings(!showAdvancedSettings)}
+              >
+                <span>
+                  {showAdvancedSettings
+                    ? "Hide advanced settings"
+                    : "Show advanced settings"}
+                </span>
+                <ChevronDown
+                  className={`ml-1 h-4 w-4 transition-transform ${showAdvancedSettings ? "rotate-180" : "rotate-0"}`}
+                />
+              </Button>
+            </div>
+          )}
+
+          {hasAdvancedSettings(currentAdapter) && showAdvancedSettings && (
             <div className="space-y-4 border-t pt-4">
               {/* baseURL */}
-              {currentAdapter !== LLMAdapter.Bedrock &&
-                currentAdapter !== LLMAdapter.Azure && (
-                  <FormField
-                    control={form.control}
-                    name="baseURL"
-                    render={({ field }) => (
-                      <FormItem>
-                        <FormLabel>API Base URL</FormLabel>
-                        <FormDescription>
-                          Leave blank to use the default base URL for the given
-                          LLM adapter.{" "}
-                          {currentAdapter === LLMAdapter.OpenAI && (
-                            <span>
-                              OpenAI default: https://api.openai.com/v1
-                            </span>
-                          )}
-                          {currentAdapter === LLMAdapter.Anthropic && (
-                            <span>
-                              Anthropic default: https://api.anthropic.com
-                              (excluding /v1/messages)
-                            </span>
-                          )}
-                        </FormDescription>
+              <FormField
+                control={form.control}
+                name="baseURL"
+                render={({ field }) => (
+                  <FormItem>
+                    <FormLabel>API Base URL</FormLabel>
+                    <FormDescription>
+                      Leave blank to use the default base URL for the given LLM
+                      adapter.{" "}
+                      {currentAdapter === LLMAdapter.OpenAI && (
+                        <span>OpenAI default: https://api.openai.com/v1</span>
+                      )}
+                      {currentAdapter === LLMAdapter.Anthropic && (
+                        <span>
+                          Anthropic default: https://api.anthropic.com
+                          (excluding /v1/messages)
+                        </span>
+                      )}
+                    </FormDescription>
 
-                        <FormControl>
-                          <Input {...field} placeholder="default" />
-                        </FormControl>
+                    <FormControl>
+                      <Input {...field} placeholder="default" />
+                    </FormControl>
 
-                        <FormMessage />
-                      </FormItem>
-                    )}
-                  />
+                    <FormMessage />
+                  </FormItem>
                 )}
+              />
 
               {/* VertexAI Location */}
               {currentAdapter === LLMAdapter.VertexAI && (
@@ -643,69 +790,8 @@ export function CreateLLMApiKeyForm({
               )}
 
               {/* Extra Headers */}
-              {currentAdapter === LLMAdapter.OpenAI ||
-              currentAdapter === LLMAdapter.Azure ? (
-                <FormField
-                  control={form.control}
-                  name="extraHeaders"
-                  render={() => (
-                    <FormItem>
-                      <FormLabel>Extra Headers</FormLabel>
-                      <FormDescription>
-                        Optional additional HTTP headers to include with
-                        requests towards LLM provider. All header values stored
-                        encrypted{" "}
-                        {isLangfuseCloud
-                          ? "on our servers"
-                          : "in your database"}
-                        .
-                      </FormDescription>
-
-                      {headerFields.map((header, index) => (
-                        <div
-                          key={header.id}
-                          className="flex flex-row space-x-2"
-                        >
-                          <Input
-                            {...form.register(`extraHeaders.${index}.key`)}
-                            placeholder="Header name"
-                          />
-                          <Input
-                            {...form.register(`extraHeaders.${index}.value`)}
-                            placeholder={
-                              mode === "update" &&
-                              existingKey?.extraHeaderKeys &&
-                              existingKey.extraHeaderKeys[index]
-                                ? "***"
-                                : "Header value"
-                            }
-                          />
-                          <Button
-                            type="button"
-                            variant="ghost"
-                            onClick={() => removeHeader(index)}
-                          >
-                            <TrashIcon className="h-4 w-4" />
-                          </Button>
-                        </div>
-                      ))}
-
-                      <Button
-                        type="button"
-                        variant="ghost"
-                        onClick={() => appendHeader({ key: "", value: "" })}
-                        className="w-full"
-                      >
-                        <PlusIcon
-                          className="-ml-0.5 mr-1.5 h-5 w-5"
-                          aria-hidden="true"
-                        />
-                        Add Header
-                      </Button>
-                    </FormItem>
-                  )}
-                />
-              ) : null}
+              {currentAdapter === LLMAdapter.OpenAI &&
+                renderExtraHeadersField()}
 
               {/* With default models */}
               <FormField
@@ -720,36 +806,11 @@ export function CreateLLMApiKeyForm({
                           Default models for the selected adapter will be
                           available in Langfuse features.
                         </FormDescription>
-                        {currentAdapter === LLMAdapter.Azure && (
-                          <FormDescription className="text-dark-yellow">
-                            Azure LLM adapter does not support default model
-                            names maintained by Langfuse. Instead, please add a
-                            custom model below that is the same as your
-                            deployment name.
-                          </FormDescription>
-                        )}
-                        {currentAdapter === LLMAdapter.Bedrock && (
-                          <FormDescription className="text-dark-yellow">
-                            Bedrock LLM adapter does not support default model
-                            names maintained by Langfuse. Instead, please add
-                            the Bedrock model IDs you have enabled in the AWS
-                            console.
-                          </FormDescription>
-                        )}
                       </span>
 
                       <FormControl>
                         <Switch
-                          disabled={
-                            currentAdapter === LLMAdapter.Azure ||
-                            currentAdapter === LLMAdapter.Bedrock
-                          }
-                          checked={
-                            currentAdapter === LLMAdapter.Azure ||
-                            currentAdapter === LLMAdapter.Bedrock
-                              ? false
-                              : field.value
-                          }
+                          checked={field.value}
                           onCheckedChange={field.onChange}
                         />
                       </FormControl>
@@ -761,64 +822,8 @@ export function CreateLLMApiKeyForm({
               />
 
               {/* Custom model names */}
-              <FormField
-                control={form.control}
-                name="customModels"
-                render={() => (
-                  <FormItem>
-                    <FormLabel>Custom models</FormLabel>
-                    <FormDescription>
-                      Custom model names accepted by given endpoint.
-                    </FormDescription>
-                    {currentAdapter === LLMAdapter.Azure && (
-                      <FormDescription className="text-dark-yellow">
-                        {
-                          "For Azure, the model name should be the same as the deployment name in Azure. For evals, choose a model with function calling capabilities."
-                        }
-                      </FormDescription>
-                    )}
-
-                    {currentAdapter === LLMAdapter.Bedrock && (
-                      <FormDescription className="text-dark-yellow">
-                        {
-                          "For Bedrock, the model name is the Bedrock Inference Profile ID, e.g. 'eu.anthropic.claude-3-5-sonnet-20240620-v1:0'"
-                        }
-                      </FormDescription>
-                    )}
-
-                    {fields.map((customModel, index) => (
-                      <span
-                        key={customModel.id}
-                        className="flex flex-row space-x-2"
-                      >
-                        <Input
-                          {...form.register(`customModels.${index}.value`)}
-                          placeholder={`Custom model name ${index + 1}`}
-                        />
-                        <Button
-                          type="button"
-                          variant="ghost"
-                          onClick={() => remove(index)}
-                        >
-                          <TrashIcon className="h-4 w-4" />
-                        </Button>
-                      </span>
-                    ))}
-                    <Button
-                      type="button"
-                      variant="ghost"
-                      onClick={() => append({ value: "" })}
-                      className="w-full"
-                    >
-                      <PlusIcon
-                        className="-ml-0.5 mr-1.5 h-5 w-5"
-                        aria-hidden="true"
-                      />
-                      Add custom model name
-                    </Button>
-                  </FormItem>
-                )}
-              />
+              {!isCustomModelsRequired(currentAdapter) &&
+                renderCustomModelsField()}
             </div>
           )}
         </DialogBody>

web/src/features/public-api/server/dailyMetrics.ts

@@ -4,6 +4,8 @@ import {
   queryClickhouse,
   TRACE_TO_OBSERVATIONS_INTERVAL,
   type DateTimeFilter,
+  getTimeframesTracesAMT,
+  measureAndReturn,
 } from "@langfuse/shared/src/server";
 
 type QueryType = {
@@ -45,7 +47,7 @@ export const generateDailyMetrics = async (props: QueryType) => {
         sum(arraySum(mapValues(mapFilter(x -> positionCaseInsensitive(x.1, 'output') > 0, o.usage_details)))) as outputUsage,
         sumMap(o.usage_details)['total'] as totalUsage,
         sum(coalesce(o.total_cost, 0)) as totalCost
-      FROM traces t FINAL
+      FROM __TRACE_TABLE__ t FINAL
       LEFT JOIN observations o FINAL on o.trace_id = t.id AND o.project_id = t.project_id
       WHERE o.project_id = {projectId: String} 
       AND t.project_id = {projectId: String}
@@ -72,7 +74,7 @@ export const generateDailyMetrics = async (props: QueryType) => {
       SELECT
         toDate(t.timestamp) as date,
         count(t.id) as countTraces
-      FROM traces t FINAL
+      FROM __TRACE_TABLE__ t FINAL
       WHERE t.project_id = {projectId: String}
       ${hasTracesFilter ? `AND ${appliedTracesFilter.query}` : ""}
       GROUP BY date
@@ -90,48 +92,104 @@ export const generateDailyMetrics = async (props: QueryType) => {
     ${props.limit !== undefined && props.page !== undefined ? `LIMIT {limit: Int32} OFFSET {offset: Int32}` : ""}
   `;
 
-  const result = await queryClickhouse<{
-    date: string;
-    countTraces: number;
-    countObservations: number;
-    totalCost: number;
-    usage: (string | null)[][];
-  }>({
-    query,
-    params: {
-      ...appliedTracesFilter.params,
-      ...appliedFilter.params,
-      projectId: props.projectId,
-      ...(props.limit !== undefined ? { limit: props.limit } : {}),
-      ...(props.page !== undefined
-        ? { offset: (props.page - 1) * props.limit }
-        : {}),
-      ...(timeFilter
-        ? {
-            cteTimeFilter: convertDateToClickhouseDateTime(timeFilter.value),
-          }
-        : {}),
+  const timestamp = props.fromTimestamp
+    ? new Date(props.fromTimestamp)
+    : timeFilter?.value;
+
+  return measureAndReturn({
+    operationName: "generateDailyMetrics",
+    projectId: props.projectId,
+    minStartTime: timestamp,
+    input: {
+      params: {
+        ...appliedTracesFilter.params,
+        ...appliedFilter.params,
+        projectId: props.projectId,
+        ...(props.limit !== undefined ? { limit: props.limit } : {}),
+        ...(props.page !== undefined
+          ? { offset: (props.page - 1) * props.limit }
+          : {}),
+        ...(timeFilter
+          ? {
+              cteTimeFilter: convertDateToClickhouseDateTime(timeFilter.value),
+            }
+          : {}),
+      },
+      tags: {
+        feature: "tracing",
+        type: "trace",
+        kind: "daily_metrics",
+        projectId: props.projectId,
+        operation_name: "generateDailyMetrics",
+      },
+      timestamp,
     },
-    clickhouseConfigs: {
-      request_timeout: 60_000, // Use 1 minute timeout for daily metrics
+    existingExecution: async (input) => {
+      const result = await queryClickhouse<{
+        date: string;
+        countTraces: number;
+        countObservations: number;
+        totalCost: number;
+        usage: (string | null)[][];
+      }>({
+        query: query.replaceAll("__TRACE_TABLE__", "traces"),
+        params: input.params,
+        tags: { ...input.tags, experiment_amt: "original" },
+        clickhouseConfigs: {
+          request_timeout: 60_000, // Use 1 minute timeout for daily metrics
+        },
+      });
+
+      return result.map((record) => ({
+        date: record.date,
+        countTraces: Number(record.countTraces),
+        countObservations: Number(record.countObservations),
+        totalCost: Number(record.totalCost),
+        usage: record.usage.map((u) => ({
+          model: u[0],
+          inputUsage: Number(u[1]),
+          outputUsage: Number(u[2]),
+          totalUsage: Number(u[3]),
+          totalCost: Number(u[4]),
+          countObservations: Number(u[5]),
+          countTraces: Number(u[6]),
+        })),
+      }));
+    },
+    newExecution: async (input) => {
+      const traceAmt = getTimeframesTracesAMT(input.timestamp);
+      const result = await queryClickhouse<{
+        date: string;
+        countTraces: number;
+        countObservations: number;
+        totalCost: number;
+        usage: (string | null)[][];
+      }>({
+        query: query.replaceAll("__TRACE_TABLE__", traceAmt),
+        params: input.params,
+        tags: { ...input.tags, experiment_amt: "new" },
+        clickhouseConfigs: {
+          request_timeout: 60_000, // Use 1 minute timeout for daily metrics
+        },
+      });
+
+      return result.map((record) => ({
+        date: record.date,
+        countTraces: Number(record.countTraces),
+        countObservations: Number(record.countObservations),
+        totalCost: Number(record.totalCost),
+        usage: record.usage.map((u) => ({
+          model: u[0],
+          inputUsage: Number(u[1]),
+          outputUsage: Number(u[2]),
+          totalUsage: Number(u[3]),
+          totalCost: Number(u[4]),
+          countObservations: Number(u[5]),
+          countTraces: Number(u[6]),
+        })),
+      }));
     },
   });
-
-  return result.map((record) => ({
-    date: record.date,
-    countTraces: Number(record.countTraces),
-    countObservations: Number(record.countObservations),
-    totalCost: Number(record.totalCost),
-    usage: record.usage.map((u) => ({
-      model: u[0],
-      inputUsage: Number(u[1]),
-      outputUsage: Number(u[2]),
-      totalUsage: Number(u[3]),
-      totalCost: Number(u[4]),
-      countObservations: Number(u[5]),
-      countTraces: Number(u[6]),
-    })),
-  }));
 };
 
 export const getDailyMetricsCount = async (props: QueryType) => {
@@ -145,16 +203,48 @@ export const getDailyMetricsCount = async (props: QueryType) => {
 
   const query = `
     SELECT count(distinct toDate(timestamp)) as count
-    FROM traces t
+    FROM __TRACE_TABLE__ t
     WHERE project_id = {projectId: String}
     ${filter.length() > 0 ? `AND ${appliedFilter.query}` : ""}
   `;
 
-  const records = await queryClickhouse<{ count: string }>({
-    query,
-    params: { ...appliedFilter.params, projectId: props.projectId },
+  const timestamp = props.fromTimestamp
+    ? new Date(props.fromTimestamp)
+    : undefined;
+
+  return measureAndReturn({
+    operationName: "getDailyMetricsCount",
+    projectId: props.projectId,
+    minStartTime: timestamp,
+    input: {
+      params: { ...appliedFilter.params, projectId: props.projectId },
+      tags: {
+        feature: "tracing",
+        type: "trace",
+        kind: "daily_metrics_count",
+        projectId: props.projectId,
+        operation_name: "getDailyMetricsCount",
+      },
+      timestamp,
+    },
+    existingExecution: async (input) => {
+      const records = await queryClickhouse<{ count: string }>({
+        query: query.replace("__TRACE_TABLE__", "traces"),
+        params: input.params,
+        tags: { ...input.tags, experiment_amt: "original" },
+      });
+      return records.map((record) => Number(record.count)).shift();
+    },
+    newExecution: async (input) => {
+      const traceAmt = getTimeframesTracesAMT(input.timestamp);
+      const records = await queryClickhouse<{ count: string }>({
+        query: query.replace("__TRACE_TABLE__", traceAmt),
+        params: input.params,
+        tags: { ...input.tags, experiment_amt: "new" },
+      });
+      return records.map((record) => Number(record.count)).shift();
+    },
   });
-  return records.map((record) => Number(record.count)).shift();
 };
 
 const filterParams = [

web/src/features/public-api/server/dataset-runs.ts

@@ -14,6 +14,19 @@ const isUniqueConstraintError = (error: any): boolean => {
   );
 };
 
+/**
+ * Create or fetch a dataset run with optimistic concurrency handling.
+ *
+ * Behavior:
+ * - First tries to find an existing run by (projectId, datasetId, name).
+ * - If not found, attempts to create it.
+ * - If creation fails due to a unique constraint (likely created concurrently),
+ *   fetches and returns the existing run.
+ * - If all steps fail, throws an error.
+ *
+ * Rationale: The public API can receive many POST requests almost simultaneously,
+ * which is not concurrency-safe without this guard.
+ */
 export const createOrFetchDatasetRun = async ({
   projectId,
   datasetId,
@@ -28,7 +41,21 @@ export const createOrFetchDatasetRun = async ({
   metadata?: Json | null;
 }) => {
   try {
-    // Attempt optimistic creation
+    // Attempt to fetch existing run
+    const existingRun = await prisma.datasetRuns.findUnique({
+      where: {
+        datasetId_projectId_name: {
+          datasetId,
+          projectId,
+          name: name,
+        },
+      },
+    });
+    if (existingRun) {
+      return existingRun;
+    }
+
+    // Attempt creation
     const datasetRun = await prisma.datasetRuns.create({
       data: {
         id: v4(),

web/src/features/public-api/server/traces.ts

@@ -84,10 +84,9 @@ export const generateTracesForPublicApi = async ({
       SELECT
         trace_id,
         project_id,
-        sum(total_cost) as total_cost,
-        date_diff('millisecond', least(min(start_time), min(end_time)), greatest(max(start_time), max(end_time))) as latency_milliseconds,
-        groupArray(id) as observation_ids
-      FROM observations FINAL
+         ${includeMetrics ? "sum(total_cost) as total_cost, date_diff('millisecond', least(min(start_time), min(end_time)), greatest(max(start_time), max(end_time))) as latency_milliseconds, " : ""}
+        groupUniqArray(id) as observation_ids
+      FROM observations ${includeMetrics ? "FINAL" : ""}
       WHERE project_id = {projectId: String}
       ${timeFilter ? `AND start_time >= {cteTimeFilter: DateTime64(3)} - ${TRACE_TO_OBSERVATIONS_INTERVAL}` : ""}
       ${environmentFilter.length() > 0 ? `AND ${appliedEnvironmentFilter.query}` : ""}
@@ -294,16 +293,48 @@ export const getTracesCountForPublicApi = async ({
 
   const query = `
     SELECT count() as count
-    FROM traces t
+    FROM __TRACE_TABLE__ t
     WHERE project_id = {projectId: String}
     ${filter.length() > 0 ? `AND ${appliedFilter.query}` : ""}
   `;
 
-  const records = await queryClickhouse<{ count: string }>({
-    query,
-    params: { ...appliedFilter.params, projectId: props.projectId },
+  const timestamp = props.fromTimestamp
+    ? new Date(props.fromTimestamp)
+    : undefined;
+
+  return measureAndReturn({
+    operationName: "getTracesCountForPublicApi",
+    projectId: props.projectId,
+    minStartTime: timestamp,
+    input: {
+      params: { ...appliedFilter.params, projectId: props.projectId },
+      tags: {
+        feature: "tracing",
+        type: "trace",
+        kind: "count",
+        projectId: props.projectId,
+        operation_name: "getTracesCountForPublicApi",
+      },
+      timestamp,
+    },
+    existingExecution: async (input) => {
+      const records = await queryClickhouse<{ count: string }>({
+        query: query.replace("__TRACE_TABLE__", "traces"),
+        params: input.params,
+        tags: { ...input.tags, experiment_amt: "original" },
+      });
+      return records.map((record) => Number(record.count)).shift();
+    },
+    newExecution: async (input) => {
+      const traceAmt = getTimeframesTracesAMT(input.timestamp);
+      const records = await queryClickhouse<{ count: string }>({
+        query: query.replace("__TRACE_TABLE__", traceAmt),
+        params: input.params,
+        tags: { ...input.tags, experiment_amt: "new" },
+      });
+      return records.map((record) => Number(record.count)).shift();
+    },
   });
-  return records.map((record) => Number(record.count)).shift();
 };
 
 const orderByColumns = [

web/src/features/public-api/types/annotation-queues.ts

@@ -57,6 +57,17 @@ export const GetAnnotationQueuesResponse = z
   })
   .strict();
 
+// POST /annotation-queues
+export const CreateAnnotationQueueBody = z
+  .object({
+    name: z.string(),
+    description: z.string().nullable(),
+    scoreConfigIds: z.array(z.string()).min(1),
+  })
+  .strict();
+
+export const CreateAnnotationQueueResponse = AnnotationQueueSchema;
+
 // GET /annotation-queues/:queueId
 export const GetAnnotationQueueByIdQuery = z
   .object({

web/src/features/query/server/queryBuilder.ts

@@ -117,10 +117,71 @@ export class QueryBuilder {
     });
   }
 
+  private validateFilters(
+    filters: z.infer<typeof queryModel>["filters"],
+    view: ViewDeclarationType,
+  ) {
+    for (const filter of filters) {
+      // Validate filters on dimension fields
+      if (filter.column in view.dimensions) {
+        const dimension = view.dimensions[filter.column];
+
+        // Array fields (like tags) validation
+        if (dimension.type === "string[]") {
+          if (filter.type === "string") {
+            throw new InvalidRequestError(
+              `Invalid filter for field '${filter.column}': Array fields require type 'arrayOptions', not 'string'. ` +
+                `Use operators like 'any of', 'all of', or 'none of' with an array of values.`,
+            );
+          }
+
+          // Additional validation: ensure value is array for arrayOptions
+          if (filter.type === "arrayOptions" && !Array.isArray(filter.value)) {
+            throw new InvalidRequestError(
+              `Invalid filter for field '${filter.column}': arrayOptions type requires an array of values, not '${typeof filter.value}'.`,
+            );
+          }
+        }
+      }
+
+      // Special validation for metadata filters
+      else if (filter.column === "metadata") {
+        if (filter.type !== "stringObject") {
+          throw new InvalidRequestError(
+            `Invalid filter for field 'metadata': Metadata filters require type 'stringObject' with a 'key' property, not '${filter.type}'. ` +
+              `Example: {"column": "metadata", "type": "stringObject", "key": "environment", "operator": "=", "value": "production"}`,
+          );
+        }
+
+        // Validate stringObject has required key
+        if (filter.type === "stringObject" && !("key" in filter)) {
+          throw new InvalidRequestError(
+            `Invalid filter for field 'metadata': stringObject type requires a 'key' property to specify which metadata field to filter on. ` +
+              `Example: {"column": "metadata", "type": "stringObject", "key": "environment", "operator": "=", "value": "production"}`,
+          );
+        }
+
+        // Validate stringObject value type
+        if (
+          filter.type === "stringObject" &&
+          typeof filter.value !== "string"
+        ) {
+          throw new InvalidRequestError(
+            // @ts-ignore
+            `Invalid filter for field 'metadata': stringObject type requires a string value, not '${typeof filter.value}'.`,
+          );
+        }
+      }
+    }
+  }
+
   private mapFilters(
     filters: z.infer<typeof queryModel>["filters"],
     view: ViewDeclarationType,
   ) {
+    // Validate all filters before processing
+    this.validateFilters(filters, view);
+
     // Transform our filters to match the column mapping format expected by createFilterFromFilterState
     const columnMappings = filters.map((filter) => {
       let clickhouseSelect: string;

web/src/features/tag/components/TagInput.tsx

@@ -9,51 +9,71 @@ type TagInputProps = React.ComponentPropsWithoutRef<
   typeof CommandPrimitive.Input
 > & {
   selectedTags: string[];
-  setSelectedTags: (tags: string[]) => void;
+  setSelectedTags?: (tags: string[]) => void;
+  allowTagRemoval?: boolean;
 };
 
 export const TagInput = React.forwardRef<
   React.ElementRef<typeof CommandPrimitive.Input>,
   TagInputProps
->(({ className, selectedTags, setSelectedTags, ...props }, ref) => {
-  const capture = usePostHogClientCapture();
-  return (
-    <div
-      className="flex flex-wrap items-center overflow-auto rounded-lg border px-2"
-      cmdk-input-wrapper=""
-    >
-      {selectedTags.length > 0 && (
-        <div className="flex flex-wrap items-center gap-x-2 gap-y-1 pt-2">
-          {selectedTags.map((tag: string) => (
-            <Button
-              key={tag}
-              variant="tertiary"
-              size="icon-sm"
-              onClick={() => {
-                const newTags = selectedTags.filter((t) => t !== tag);
-                setSelectedTags(newTags);
-                capture("tag:remove_tag", {
-                  name: tag,
-                });
-              }}
-            >
-              {tag}
-              <X className="ml-1 h-3 w-3" />
-            </Button>
-          ))}
-        </div>
-      )}
-      <CommandPrimitive.Input
-        ref={ref}
-        className={cn(
-          "placeholder:muted-foreground flex h-8 w-full rounded-md border-transparent bg-transparent px-1 text-sm outline-none focus:border-0 focus:border-none focus:border-transparent focus:ring-0 disabled:cursor-not-allowed disabled:opacity-50",
-          className,
+>(
+  (
+    {
+      className,
+      selectedTags,
+      setSelectedTags,
+      allowTagRemoval = false,
+      ...props
+    },
+    ref,
+  ) => {
+    const capture = usePostHogClientCapture();
+
+    const removeTag = (tagToRemove: string) => {
+      if (setSelectedTags && allowTagRemoval) {
+        setSelectedTags(selectedTags.filter((t) => t !== tagToRemove));
+        capture("tag:remove_tag", {
+          name: tagToRemove,
+        });
+      }
+    };
+
+    return (
+      <div
+        className="flex flex-wrap items-center overflow-auto rounded-lg border px-2"
+        cmdk-input-wrapper=""
+      >
+        {selectedTags.length > 0 && (
+          <div className="flex flex-wrap items-center gap-x-2 gap-y-1 pt-2">
+            {selectedTags.map((tag: string) => (
+              <Button
+                key={tag}
+                variant="tertiary"
+                size="icon-sm"
+                disabled={!allowTagRemoval}
+                className={
+                  allowTagRemoval ? "cursor-pointer" : "cursor-default"
+                }
+                onClick={allowTagRemoval ? () => removeTag(tag) : undefined}
+              >
+                {tag}
+                {allowTagRemoval && <X className="ml-1 h-3 w-3" />}
+              </Button>
+            ))}
+          </div>
         )}
-        autoFocus
-        {...props}
-      />
-    </div>
-  );
-});
+        <CommandPrimitive.Input
+          ref={ref}
+          className={cn(
+            "placeholder:muted-foreground flex h-8 w-full rounded-md border-transparent bg-transparent px-1 text-sm outline-none focus:border-0 focus:border-none focus:border-transparent focus:ring-0 disabled:cursor-not-allowed disabled:opacity-50",
+            className,
+          )}
+          autoFocus
+          {...props}
+        />
+      </div>
+    );
+  },
+);
 
 TagInput.displayName = CommandPrimitive.Input.displayName;

web/src/features/tag/components/TagManager.tsx

@@ -22,6 +22,7 @@ type TagManagerProps = {
   mutateTags: (value: string[]) => void;
   className?: string;
   isTableCell?: boolean;
+  allowTagRemoval?: boolean;
 };
 
 const TagManager = ({
@@ -33,6 +34,7 @@ const TagManager = ({
   mutateTags,
   className,
   isTableCell = false,
+  allowTagRemoval = true,
 }: TagManagerProps) => {
   const {
     selectedTags,
@@ -114,6 +116,7 @@ const TagManager = ({
             onValueChange={setInputValue}
             selectedTags={selectedTags}
             setSelectedTags={setSelectedTags}
+            allowTagRemoval={allowTagRemoval}
           />
           <CommandList>
             <CommandGroup

web/src/features/tag/components/TagPromptDetailsPopover.tsx

@@ -2,7 +2,7 @@ import React, { useState } from "react";
 import { api } from "@/src/utils/api";
 import { useHasProjectAccess } from "@/src/features/rbac/utils/checkProjectAccess";
 import { type RouterOutput } from "@/src/utils/types";
-import TagManager from "@/src/features/tag/components/TagMananger";
+import TagManager from "@/src/features/tag/components/TagManager";
 import { trpcErrorToast } from "@/src/utils/trpcErrorToast";
 
 type TagPromptDetailsPopoverProps = {

web/src/features/tag/components/TagPromptPopover.tsx

@@ -2,7 +2,7 @@ import React, { useState } from "react";
 import { api } from "@/src/utils/api";
 import { useHasProjectAccess } from "@/src/features/rbac/utils/checkProjectAccess";
 import { type RouterOutput, type RouterInput } from "@/src/utils/types";
-import TagManager from "@/src/features/tag/components/TagMananger";
+import TagManager from "@/src/features/tag/components/TagManager";
 import { trpcErrorToast } from "@/src/utils/trpcErrorToast";
 
 type TagPromptPopverProps = {

web/src/features/tag/components/TagTraceDetailsPopover.tsx

@@ -2,7 +2,7 @@ import React, { useState } from "react";
 import { api } from "@/src/utils/api";
 import { useHasProjectAccess } from "@/src/features/rbac/utils/checkProjectAccess";
 import { type RouterOutput } from "@/src/utils/types";
-import TagManager from "@/src/features/tag/components/TagMananger";
+import TagManager from "@/src/features/tag/components/TagManager";
 import { trpcErrorToast } from "@/src/utils/trpcErrorToast";
 
 type TagTraceDetailsPopoverProps = {
@@ -85,6 +85,7 @@ export function TagTraceDetailsPopover({
       isLoading={isLoading}
       mutateTags={mutateTags}
       className={className}
+      allowTagRemoval={false}
     />
   );
 }

web/src/features/tag/components/TagTracePopver.tsx

@@ -2,7 +2,7 @@ import React, { useState } from "react";
 import { api } from "@/src/utils/api";
 import { useHasProjectAccess } from "@/src/features/rbac/utils/checkProjectAccess";
 import { type RouterOutput, type RouterInput } from "@/src/utils/types";
-import TagManager from "@/src/features/tag/components/TagMananger";
+import TagManager from "@/src/features/tag/components/TagManager";
 import { trpcErrorToast } from "@/src/utils/trpcErrorToast";
 
 type TagTracePopoverProps = {
@@ -75,6 +75,7 @@ export function TagTracePopover({
       mutateTags={mutateTags}
       className={className}
       isTableCell
+      allowTagRemoval={false}
     />
   );
 }

web/src/pages/api/admin/bullmq/index.ts

@@ -56,13 +56,6 @@ export default async function handler(
       return;
     }
 
-    const body = ManageBullBody.safeParse(req.body);
-
-    if (!body.success) {
-      res.status(400).json({ error: body.error });
-      return;
-    }
-
     if (req.method === "GET") {
       const queues: string[] = Object.values(QueueName);
       queues.push(...IngestionQueue.getShardNames());
@@ -94,6 +87,13 @@ export default async function handler(
       return res.status(200).json(queueCounts);
     }
 
+    const body = ManageBullBody.safeParse(req.body);
+
+    if (!body.success) {
+      res.status(400).json({ error: body.error });
+      return;
+    }
+
     if (req.method === "POST" && body.data.action === "remove") {
       logger.info(
         `Removing jobs for queues ${body.data.queueNames.join(", ")}`,

web/src/pages/api/public/annotation-queues/index.ts

@@ -2,12 +2,14 @@ import { prisma } from "@langfuse/shared/src/db";
 import { withMiddlewares } from "@/src/features/public-api/server/withMiddlewares";
 import { createAuthedProjectAPIRoute } from "@/src/features/public-api/server/createAuthedProjectAPIRoute";
 import {
+  CreateAnnotationQueueBody,
+  CreateAnnotationQueueResponse,
   GetAnnotationQueuesQuery,
   GetAnnotationQueuesResponse,
 } from "@/src/features/public-api/types/annotation-queues";
+import { InvalidRequestError, MethodNotAllowedError } from "@langfuse/shared";
 
 export default withMiddlewares({
-  // NOTE: Post API requires entitlement check
   GET: createAuthedProjectAPIRoute({
     name: "Get annotation queues",
     querySchema: GetAnnotationQueuesQuery,
@@ -54,4 +56,72 @@ export default withMiddlewares({
       };
     },
   }),
+
+  POST: createAuthedProjectAPIRoute({
+    name: "Create annotation queue",
+    bodySchema: CreateAnnotationQueueBody,
+    responseSchema: CreateAnnotationQueueResponse,
+    fn: async ({ body, auth }) => {
+      // entitlement check
+      if (auth.scope.plan === "cloud:hobby") {
+        if (
+          (await prisma.annotationQueue.count({
+            where: {
+              projectId: auth.scope.projectId,
+            },
+          })) >= 1
+        ) {
+          throw new MethodNotAllowedError(
+            "Maximum number of annotation queues reached on Hobby plan.",
+          );
+        }
+      }
+
+      const existingQueue = await prisma.annotationQueue.findFirst({
+        where: {
+          projectId: auth.scope.projectId,
+          name: body.name,
+        },
+      });
+
+      if (existingQueue) {
+        throw new InvalidRequestError("A queue with this name already exists.");
+      }
+
+      // verify the score configs exist
+      const scoreConfigs = await prisma.scoreConfig.findMany({
+        where: {
+          id: { in: body.scoreConfigIds },
+          projectId: auth.scope.projectId,
+        },
+        select: {
+          id: true,
+        },
+      });
+      const scoreConfigIdSet = new Set(scoreConfigs.map((config) => config.id));
+      if (body.scoreConfigIds.some((id) => !scoreConfigIdSet.has(id))) {
+        throw new InvalidRequestError(
+          "At least one of the score config IDs cannot be found for the given project.",
+        );
+      }
+
+      const queue = await prisma.annotationQueue.create({
+        data: {
+          projectId: auth.scope.projectId,
+          name: body.name,
+          description: body.description,
+          scoreConfigIds: body.scoreConfigIds,
+        },
+      });
+
+      return {
+        id: queue.id,
+        name: queue.name,
+        description: queue.description,
+        scoreConfigIds: queue.scoreConfigIds,
+        createdAt: queue.createdAt,
+        updatedAt: queue.updatedAt,
+      };
+    },
+  }),
 });

web/src/pages/api/public/health.ts

@@ -5,6 +5,7 @@ import { prisma } from "@langfuse/shared/src/db";
 import {
   convertDateToClickhouseDateTime,
   logger,
+  measureAndReturn,
   queryClickhouse,
   traceException,
 } from "@langfuse/shared/src/server";
@@ -37,20 +38,45 @@ export default async function handler(
     try {
       if (failIfNoRecentEvents) {
         const now = new Date();
-        const traces = await queryClickhouse({
-          query: `
-            SELECT id
-            FROM traces
-            WHERE timestamp <= {now: DateTime64(3)}
-            AND timestamp >= {now: DateTime64(3)} - INTERVAL 3 MINUTE
-            LIMIT 1
-          `,
-          params: {
+        const traces = await measureAndReturn({
+          operationName: "healthCheckTraces",
+          projectId: "__CROSS_PROJECT__",
+          input: {
             now: convertDateToClickhouseDateTime(now),
           },
-          tags: {
-            feature: "health-check",
-            type: "trace",
+          existingExecution: async (input: { now: string }) => {
+            return queryClickhouse<{ id: string }>({
+              query: `
+                SELECT id
+                FROM traces
+                WHERE timestamp <= {now: DateTime64(3)}
+                AND timestamp >= {now: DateTime64(3)} - INTERVAL 3 MINUTE
+                LIMIT 1
+              `,
+              params: input,
+              tags: {
+                feature: "health-check",
+                type: "trace",
+                experiment_amt: "original",
+              },
+            });
+          },
+          newExecution: async (input: { now: string }) => {
+            return queryClickhouse<{ id: string }>({
+              query: `
+                SELECT id
+                FROM traces_7d_amt
+                WHERE start_time <= {now: DateTime64(3)}
+                AND start_time >= {now: DateTime64(3)} - INTERVAL 3 MINUTE
+                LIMIT 1
+              `,
+              params: input,
+              tags: {
+                feature: "health-check",
+                type: "trace",
+                experiment_amt: "new",
+              },
+            });
           },
         });
         const observations = await queryClickhouse({

web/src/server/api/routers/public.ts

@@ -2,7 +2,6 @@ import { VERSION } from "@/src/constants/VERSION";
 import { env } from "@/src/env.mjs";
 import { createTRPCRouter, publicProcedure } from "@/src/server/api/trpc";
 import { logger } from "@langfuse/shared/src/server";
-import { TRPCError } from "@trpc/server";
 import { z } from "zod/v4";
 
 const versionSchema = z.string().regex(/^v\d+\.\d+\.\d+(?:[-+].+)?$/); // e.g. v1.2.3, v1.2.3-rc.1, v1.2.3+build.123
@@ -79,27 +78,33 @@ export const publicRouter = createTRPCRouter({
       );
       body = await response.json();
     } catch (error) {
-      logger.info(
+      logger.error(
         "[trpc.public.checkUpdate] failed to fetch latest-release api",
+        {
+          error,
+        },
       );
       return null;
     }
 
     const releases = ReleaseApiRes.safeParse(body);
     if (!releases.success) {
-      throw new TRPCError({
-        code: "INTERNAL_SERVER_ERROR",
-        message: "Release API response is invalid",
-      });
+      logger.error(
+        "[trpc.public.checkUpdate] Release API response is invalid, does not match schema",
+        {
+          error: releases.error,
+        },
+      );
+      return null;
     }
     const langfuseRelease = releases.data.find(
       (release) => release.repo === "langfuse/langfuse",
     );
     if (!langfuseRelease) {
-      throw new TRPCError({
-        code: "INTERNAL_SERVER_ERROR",
-        message: "Release API response is invalid",
-      });
+      logger.error(
+        "[trpc.public.checkUpdate] Release API response is invalid, does not contain langfuse/langfuse",
+      );
+      return null;
     }
 
     const updateType = compareVersions(VERSION, langfuseRelease.latestRelease);

worker/package.json

@@ -1,6 +1,6 @@
 {
   "name": "worker",
-  "version": "3.96.2",
+  "version": "3.97.4",
   "description": "",
   "license": "MIT",
   "private": true,

worker/src/__tests__/experimentsService.test.ts

@@ -588,26 +588,7 @@ describe("create experiment job calls with langfuse server side tracing", async
     await createExperimentJobPostgres({ event: mockEvent });
 
     // Verify callLLM was called with correct trace parameters
-    expect(callLLM).toHaveBeenCalledWith(
-      expect.any(Object),
-      expect.any(Array),
-      expect.any(Object),
-      expect.any(String),
-      expect.any(String),
-      expect.objectContaining({
-        environment: PROMPT_EXPERIMENT_ENVIRONMENT,
-        traceName: expect.stringMatching(/^dataset-run-item-/),
-        traceId: expect.any(String),
-        projectId: mockEvent.projectId,
-        authCheck: expect.objectContaining({
-          validKey: true,
-          scope: expect.objectContaining({
-            projectId: mockEvent.projectId,
-            accessLevel: "project",
-          }),
-        }),
-      }),
-    );
+    expect(callLLM).toHaveBeenCalledTimes(0);
   });
 });
 

worker/src/backgroundMigrations/migrateTracesToTracesAMTs.ts

@@ -16,6 +16,7 @@ type MigrationState = {
   maxDate: string | undefined;
   minDate: string | undefined;
   queryTimeoutMinutes: number | undefined;
+  targetTracesAllAmtOnly: boolean | undefined;
 };
 
 /**
@@ -35,6 +36,7 @@ async function checkQueryExists(
       SELECT COUNT(*) > 0 AS exists
       FROM ${queryLogTable}
       WHERE query_id = '${queryId}'
+      ${env.CLICKHOUSE_CLUSTER_ENABLED === "true" ? " SETTINGS skip_unavailable_shards = 1 " : ""}
     `,
     format: "JSONEachRow",
   });
@@ -60,6 +62,7 @@ async function checkCompletedQuery(
       FROM ${queryLogTable}
       WHERE query_id = '${queryId}' AND type != 'QueryStart'
       LIMIT 1
+      ${env.CLICKHOUSE_CLUSTER_ENABLED === "true" ? " SETTINGS skip_unavailable_shards = 1 " : ""}
     `,
     format: "JSONEachRow",
   });
@@ -89,9 +92,7 @@ async function executeLongRunningQuery(
   const abortController = new AbortController();
   const timeoutMs = timeoutMinutes * 60 * 1000;
 
-  logger.info(
-    `[Background Migration] Executing traces_null backfill query ${queryId}`,
-  );
+  logger.info(`[Background Migration] Executing backfill query ${queryId}`);
 
   // Start the query execution
   const queryPromise = client.command({
@@ -206,7 +207,7 @@ export default class MigrateTracesToTracesAMTs implements IBackgroundMigration {
       };
     }
 
-    // Check if new ClickHouse tables exists
+    // Check if required ClickHouse tables exist
     const tables = await clickhouseClient().query({
       query: "SHOW TABLES",
     });
@@ -247,6 +248,10 @@ export default class MigrateTracesToTracesAMTs implements IBackgroundMigration {
     const queryTimeoutMinutes =
       initialMigrationState.state?.queryTimeoutMinutes ??
       (args.queryTimeoutMinutes as number | undefined);
+    const targetTracesAllAmtOnly =
+      initialMigrationState.state?.targetTracesAllAmtOnly ??
+      (args.targetTracesAllAmtOnly as boolean | undefined) ??
+      false;
 
     const maxDate = initialMigrationState.state?.maxDate
       ? new Date(initialMigrationState.state.maxDate)
@@ -262,6 +267,7 @@ export default class MigrateTracesToTracesAMTs implements IBackgroundMigration {
           maxDate,
           minDate,
           queryTimeoutMinutes,
+          targetTracesAllAmtOnly,
         },
       },
     });
@@ -270,14 +276,21 @@ export default class MigrateTracesToTracesAMTs implements IBackgroundMigration {
       const queryStart = Date.now();
 
       // @ts-ignore
-      const migrationState: { state: { maxDate: string; minDate: string } } =
-        await prisma.backgroundMigration.findUniqueOrThrow({
-          where: { id: backgroundMigrationId },
-          select: { state: true },
-        });
+      const migrationState: {
+        state: {
+          maxDate: string;
+          minDate: string;
+          targetTracesAllAmtOnly?: boolean;
+        };
+      } = await prisma.backgroundMigration.findUniqueOrThrow({
+        where: { id: backgroundMigrationId },
+        select: { state: true },
+      });
 
       const maxDate = new Date(migrationState.state.maxDate);
       const minDate = new Date(migrationState.state.minDate);
+      const targetTracesAllAmtOnly =
+        migrationState.state.targetTracesAllAmtOnly ?? false;
 
       // Get current month in YYYYMM format
       const currentMonth = maxDate.toISOString().slice(0, 7).replace("-", "");
@@ -285,45 +298,89 @@ export default class MigrateTracesToTracesAMTs implements IBackgroundMigration {
         `[Background Migration] Migrating traces for ${currentMonth}`,
       );
 
-      const query = `
-        INSERT INTO traces_null
-        SELECT 
-          -- Identifiers
-          project_id,
-          id,
-          timestamp as start_time,
-          null as end_time,
-          name,
-          
-          -- Metadata properties
-          metadata,
-          user_id,
-          session_id,
-          environment,
-          tags,
-          version, 
-          release,
-          
-          -- UI Properties
-          bookmarked,
-          public,
-          
-          -- Aggregations (ignored)
-          [] as observation_ids,
-          [] as score_ids,
-          map() as cost_details,
-          map() as usage_details,
-          
-          -- Input/Output
-          input,
-          output,
-          
-          created_at,
-          updated_at,
-          event_ts
-        FROM traces
-        WHERE toYYYYMM(timestamp) = ${currentMonth}
-      `;
+      const targetTable = targetTracesAllAmtOnly
+        ? "traces_all_amt"
+        : "traces_null";
+
+      const query = targetTracesAllAmtOnly
+        ? `
+          INSERT INTO ${targetTable}
+          SELECT
+            -- Identifiers
+            project_id,
+            id,
+            t.timestamp as timestamp,
+            t.timestamp as start_time,
+            t.timestamp as end_time,
+            name,
+
+            -- Metadata properties
+            metadata,
+            user_id,
+            session_id,
+            environment,
+            tags,
+            version,
+            release,
+
+            -- UI Properties
+            arrayReduce('argMaxState', [toNullable(bookmarked)], [event_ts]) as bookmarked,
+            arrayReduce('argMaxState', [toNullable(public)], [event_ts]) as public,
+
+            -- Aggregations
+            [] as observation_ids,
+            [] as score_ids,
+            map() as cost_details,
+            map() as usage_details,
+
+            -- Input/Output
+            arrayReduce('argMaxState', [coalesce(input, '')], [if(coalesce(input, '') <> '', event_ts, toDateTime64(0, 3))]) as input,
+            arrayReduce('argMaxState', [coalesce(output, '')], [if(coalesce(output, '') <> '', event_ts, toDateTime64(0, 3))]) as output,
+
+            created_at,
+            updated_at
+          FROM traces t
+          WHERE toYYYYMM(t.timestamp) = ${currentMonth}
+        `
+        : `
+          INSERT INTO ${targetTable}
+          SELECT
+            -- Identifiers
+            project_id,
+            id,
+            timestamp as start_time,
+            null as end_time,
+            name,
+
+            -- Metadata properties
+            metadata,
+            user_id,
+            session_id,
+            environment,
+            tags,
+            version,
+            release,
+
+            -- UI Properties
+            bookmarked,
+            public,
+
+            -- Aggregations (ignored)
+            [] as observation_ids,
+            [] as score_ids,
+            map() as cost_details,
+            map() as usage_details,
+
+            -- Input/Output
+            input,
+            output,
+
+            created_at,
+            updated_at,
+            event_ts
+          FROM traces
+          WHERE toYYYYMM(timestamp) = ${currentMonth}
+        `;
 
       await executeLongRunningQuery(query, queryTimeoutMinutes ?? 90);
 
@@ -339,7 +396,7 @@ export default class MigrateTracesToTracesAMTs implements IBackgroundMigration {
       });
 
       logger.info(
-        `[Background Migration] Inserted traces into traces_null for ${currentMonth} in ${Date.now() - queryStart}ms`,
+        `[Background Migration] Inserted traces into ${targetTable} for ${currentMonth} in ${Date.now() - queryStart}ms`,
       );
 
       if (maxDate < minDate) {
@@ -410,6 +467,10 @@ async function main() {
         short: "t",
         default: "90",
       },
+      targetTracesAllAmtOnly: {
+        type: "boolean",
+        default: false,
+      },
     },
   });
 

worker/src/constants/VERSION.ts

@@ -1 +1 @@
-export const VERSION = "v3.96.2";
+export const VERSION = "v3.97.4";

worker/src/constants/default-model-prices.json

@@ -1873,8 +1873,8 @@
   },
   {
     "id": "cmcnjkrfa000207l4fpnh5mnv",
-    "model_name": "gemini-2.5-flash-lite-preview-06-17",
-    "match_pattern": "(?i)^(gemini-2.5-flash-lite-preview-06-17)$",
+    "model_name": "gemini-2.5-flash-lite",
+    "match_pattern": "(?i)^(gemini-2.5-flash-lite)$",
     "created_at": "2025-07-03T13:44:06.964Z",
     "updated_at": "2025-07-03T13:44:06.964Z",
     "prices": {
@@ -1932,6 +1932,27 @@
     },
     "tokenizer_id": "openai"
   },
+  {
+    "id": "12543803-2d5f-4189-addc-821ad71c8b55",
+    "model_name": "gpt-5-2025-08-07",
+    "match_pattern": "(?i)^(gpt-5-2025-08-07)$",
+    "created_at": "2025-08-11T08:00:00.000Z",
+    "updated_at": "2025-08-11T08:00:00.000Z",
+    "prices": {
+      "input": 1.25e-6,
+      "input_cached_tokens": 0.125e-6,
+      "output": 10.0e-6,
+      "input_cache_read": 0.125e-6,
+      "output_reasoning_tokens": 10e-6,
+      "output_reasoning": 10e-6
+    },
+    "tokenizer_config": {
+      "tokensPerName": 1,
+      "tokenizerModel": "gpt-4",
+      "tokensPerMessage": 3
+    },
+    "tokenizer_id": "openai"
+  },
   {
     "id": "3d6a975a-a42d-4ea2-a3ec-4ae567d5a364",
     "model_name": "gpt-5-mini",
@@ -1953,6 +1974,27 @@
     },
     "tokenizer_id": "openai"
   },
+  {
+    "id": "03b83894-7172-4e1e-8e8b-37d792484efd",
+    "model_name": "gpt-5-mini-2025-08-07",
+    "match_pattern": "(?i)^(gpt-5-mini-2025-08-07)$",
+    "created_at": "2025-08-11T08:00:00.000Z",
+    "updated_at": "2025-08-11T08:00:00.000Z",
+    "prices": {
+      "input": 0.25e-6,
+      "input_cached_tokens": 0.025e-6,
+      "output": 2.0e-6,
+      "input_cache_read": 0.025e-6,
+      "output_reasoning_tokens": 2e-6,
+      "output_reasoning": 2e-6
+    },
+    "tokenizer_config": {
+      "tokensPerName": 1,
+      "tokenizerModel": "gpt-4",
+      "tokensPerMessage": 3
+    },
+    "tokenizer_id": "openai"
+  },
   {
     "id": "f0b40234-b694-4c40-9494-7b0efd860fb9",
     "model_name": "gpt-5-nano",
@@ -1974,6 +2016,27 @@
     },
     "tokenizer_id": "openai"
   },
+  {
+    "id": "4489fde4-a594-4011-948b-526989300cd3",
+    "model_name": "gpt-5-nano-2025-08-07",
+    "match_pattern": "(?i)^(gpt-5-nano-2025-08-07)$",
+    "created_at": "2025-08-11T08:00:00.000Z",
+    "updated_at": "2025-08-11T08:00:00.000Z",
+    "prices": {
+      "input": 0.05e-6,
+      "input_cached_tokens": 0.005e-6,
+      "output": 0.4e-6,
+      "input_cache_read": 0.005e-6,
+      "output_reasoning_tokens": 0.4e-6,
+      "output_reasoning": 0.4e-6
+    },
+    "tokenizer_config": {
+      "tokensPerName": 1,
+      "tokenizerModel": "gpt-4",
+      "tokensPerMessage": 3
+    },
+    "tokenizer_id": "openai"
+  },
   {
     "id": "8ba72ee3-ebe8-4110-a614-bf81094447e5",
     "model_name": "gpt-5-chat-latest",

worker/src/ee/dataRetention/handleDataRetentionProcessingJob.ts

@@ -56,22 +56,21 @@ export const handleDataRetentionProcessingJob = async (job: Job) => {
     );
   }
 
-  await removeIngestionEventsFromS3AndDeleteClickhouseRefsForProject(
-    projectId,
-    cutoffDate,
-  );
-
   // Delete ClickHouse (TTL / Delete Queries)
   logger.info(
-    `[Data Retention] Deleting ClickHouse data older than ${retention} days for project ${projectId}`,
+    `[Data Retention] Deleting ClickHouse and S3 data older than ${retention} days for project ${projectId}`,
   );
   await Promise.all([
+    removeIngestionEventsFromS3AndDeleteClickhouseRefsForProject(
+      projectId,
+      cutoffDate,
+    ),
     deleteTracesOlderThanDays(projectId, cutoffDate),
     deleteObservationsOlderThanDays(projectId, cutoffDate),
     deleteScoresOlderThanDays(projectId, cutoffDate),
   ]);
   logger.info(
-    `[Data Retention] Deleted ClickHouse data older than ${retention} days for project ${projectId}`,
+    `[Data Retention] Deleted ClickHouse and S3 data older than ${retention} days for project ${projectId}`,
   );
 
   // Set S3 Lifecycle for deletion (Future)

worker/src/env.ts

@@ -262,6 +262,9 @@ const EnvSchema = z.object({
     .positive()
     .default(2),
   LANGFUSE_DELETE_BATCH_SIZE: z.coerce.number().positive().default(2000),
+  LANGFUSE_EXPERIMENT_DATASET_RUN_ITEMS_TRACE_SOURCE_CH: z
+    .enum(["true", "false"])
+    .default("true"),
 });
 
 export const env: z.infer<typeof EnvSchema> =

worker/src/features/database-read-stream/getDatabaseReadStream.ts

@@ -135,25 +135,46 @@ export const getDatabaseReadStream = async ({
             clickhouseConfigs,
           });
 
-          return scores.map((score) => ({
-            id: score.id,
-            traceId: score.traceId,
-            sessionId: score.sessionId,
-            datasetRunId: score.datasetRunId,
-            timestamp: score.timestamp,
-            source: score.source,
-            name: score.name,
-            dataType: score.dataType,
-            value: score.value,
-            stringValue: score.stringValue,
-            comment: score.comment,
-            metadata: score.metadata,
-            observationId: score.observationId,
-            traceName: score.traceName,
-            userId: score.traceUserId,
-            traceTags: score.traceTags,
-            environment: score.environment,
-          }));
+          // Get author user info for scores
+          // Only users that have valid project write access may write scores
+          // Only users with at least MEMBER permissions (projectMembers:read) may trigger batch exports
+          const users = await prisma.user.findMany({
+            where: {
+              id: {
+                in: scores
+                  .map((score) => score.authorUserId)
+                  .filter((s): s is string => Boolean(s)),
+              },
+            },
+            select: {
+              id: true,
+              name: true,
+            },
+          });
+
+          return scores.map((score) => {
+            const user = users.find((u) => u.id === score.authorUserId);
+            return {
+              id: score.id,
+              traceId: score.traceId,
+              sessionId: score.sessionId,
+              datasetRunId: score.datasetRunId,
+              timestamp: score.timestamp,
+              source: score.source,
+              name: score.name,
+              dataType: score.dataType,
+              value: score.value,
+              stringValue: score.stringValue,
+              comment: score.comment,
+              metadata: score.metadata,
+              observationId: score.observationId,
+              traceName: score.traceName,
+              userId: score.traceUserId,
+              traceTags: score.traceTags,
+              environment: score.environment,
+              authorUserName: user?.name ?? null,
+            };
+          });
         },
         env.BATCH_EXPORT_PAGE_SIZE,
         rowLimit,

worker/src/features/evaluation/evalService.ts

@@ -34,7 +34,6 @@ import {
 } from "./traceFilterUtils";
 import {
   ChatMessageRole,
-  ForbiddenError,
   LangfuseNotFoundError,
   Prisma,
   singleFilter,
@@ -490,13 +489,7 @@ export const evaluate = async ({
     return;
   }
 
-  if (!job?.job_input_trace_id) {
-    throw new ForbiddenError(
-      "Jobs can only be executed on traces and dataset runs for now.",
-    );
-  }
-
-  if (job.status === "CANCELLED") {
+  if (job.status === "CANCELLED" || !job?.job_input_trace_id) {
     logger.debug(`Job ${job.id} for project ${event.projectId} was cancelled.`);
 
     await kyselyPrisma.$kysely

worker/src/features/experiments/utils.ts

@@ -29,6 +29,7 @@ import {
 import { kyselyPrisma, prisma } from "@langfuse/shared/src/db";
 import z from "zod/v4";
 import { createHash } from "crypto";
+import { env } from "../../env";
 
 export enum TraceExecutionSource {
   // eslint-disable-next-line no-unused-vars
@@ -52,7 +53,9 @@ export enum TraceExecutionSource {
  *
  */
 export const shouldCreateTrace = (source: TraceExecutionSource) => {
-  // TODO: use env variable instead
+  if (env.LANGFUSE_EXPERIMENT_DATASET_RUN_ITEMS_TRACE_SOURCE_CH === "true") {
+    return source === TraceExecutionSource.CLICKHOUSE;
+  }
   return source === TraceExecutionSource.POSTGRES;
 };
 

worker/src/features/traces/processPostgresTraceDelete.ts

@@ -9,21 +9,13 @@ export const processPostgresTraceDelete = async (
     `Deleting traces ${JSON.stringify(traceIds)} in project ${projectId} from Postgres`,
   );
   try {
-    await prisma.jobExecution.updateMany({
+    await prisma.jobExecution.deleteMany({
       where: {
         jobInputTraceId: {
           in: traceIds,
         },
         projectId: projectId,
       },
-      data: {
-        jobInputTraceId: {
-          set: null,
-        },
-        jobInputObservationId: {
-          set: null,
-        },
-      },
     });
   } catch (e) {
     logger.error(

worker/src/scripts/upsertDefaultModelPrices.ts

@@ -1,7 +1,7 @@
 import { z } from "zod/v4";
 import { prisma } from "@langfuse/shared/src/db";
 import defaultModelPrices from "../constants/default-model-prices.json";
-import { logger } from "@langfuse/shared/src/server";
+import { clearFullModelCache, logger } from "@langfuse/shared/src/server";
 
 const DefaultModelPriceSchema = z.object({
   id: z.string(),
@@ -50,6 +50,7 @@ const ExistingModelPriceSchema = z.object({
 export const upsertDefaultModelPrices = async (force = false) => {
   const startTime = Date.now();
   try {
+    let hasUpdates = false;
     logger.debug(`Starting upsert of default model prices (force = ${force})`);
 
     const parsedDefaultModelPrices = z
@@ -129,6 +130,17 @@ export const upsertDefaultModelPrices = async (force = false) => {
           continue;
         }
 
+        if (
+          !existingModelUpdateDate &&
+          Object.keys(defaultModelPrice.prices).length === 0
+        ) {
+          logger.debug(
+            `No new and existing prices for ${defaultModelPrice.model_name} (${defaultModelPrice.id}). Skipping.`,
+          );
+
+          continue;
+        }
+
         // Upsert model and prices in a transaction
         promises.push(
           prisma
@@ -205,10 +217,19 @@ export const upsertDefaultModelPrices = async (force = false) => {
         );
       }
 
+      if (promises.length > 0) {
+        hasUpdates = true;
+      }
+
       await Promise.all(promises);
+
       logger.debug(`Completed batch ${i + 1} of ${numBatches}`);
     }
 
+    if (hasUpdates) {
+      await clearFullModelCache();
+    }
+
     logger.info(
       `Finished upserting default model prices in ${Date.now() - startTime}ms`,
     );

worker/src/services/IngestionService/README.md

@@ -502,16 +502,16 @@ This checklist documents all references and invocations to the `traces` table gr
 
 ### 4. Aggregation and Analytics Queries
 
-- [ ] **getTracesCountForPublicApi()** - `web/src/features/public-api/server/traces.ts:299`
-- [ ] **generateDailyMetrics()** - `web/src/features/public-api/server/dailyMetrics.ts:93`
-- [ ] **getDailyMetricsCount()** - `web/src/features/public-api/server/dailyMetrics.ts:153`
+- [x] **getTracesCountForPublicApi()** - `web/src/features/public-api/server/traces.ts:299`
+- [x] **generateDailyMetrics()** - `web/src/features/public-api/server/dailyMetrics.ts:93`
+- [x] **getDailyMetricsCount()** - `web/src/features/public-api/server/dailyMetrics.ts:153`
 - [x] **generateObservationsForPublicApi()** - `web/src/features/public-api/server/observations.ts:80`
 - [x] **getObservationsCountForPublicApi()** - `web/src/features/public-api/server/observations.ts:108`
 - [x] **getObservationsTableInternal()** - `packages/shared/src/server/repositories/observations.ts:565`
 - [x] **_handleGenerateScoresForPublicApi()** - `web/src/features/public-api/server/scores.ts:101`
 - [x] **_handleGetScoresCountForPublicApi()** - `web/src/features/public-api/server/scores.ts:181`
 - [x] **getScoresUiGeneric()** - `packages/shared/src/server/repositories/scores.ts:825`
-- [ ] **getNumericScoreHistogram()** - `packages/shared/src/server/repositories/scores.ts:1074`
+- [x] **getNumericScoreHistogram()** - `packages/shared/src/server/repositories/scores.ts:1074`
 - [x] **getTracesGroupedByName()** - `packages/shared/src/server/repositories/traces.ts:489-535`
 - [x] **getTracesGroupedByUsers()** - `packages/shared/src/server/repositories/traces.ts:537-597`
 - [x] **getTracesGroupedByTags()** - `packages/shared/src/server/repositories/traces.ts:605-640`
@@ -527,10 +527,10 @@ Note: The measureAndReturn utility does not handle query streams well as it prom
 We need to cover these queries manually and cannot run a comparison.
 We could use an opt-in on a projectId basis.
 
-- [ ] **getTracesForPostHog()** - `packages/shared/src/server/repositories/traces.ts:1026-1113`
-- [ ] **getScoresForPostHog()** - `packages/shared/src/server/repositories/scores.ts:1328`
-- [ ] **getGenerationsForPosthog()** - `packages/shared/src/server/repositories/observations.ts:1481`
-- [ ] **getTracesForBlobStorageExport()** - `packages/shared/src/server/repositories/traces.ts:980-1024`
+- [x] **getTracesForPostHog()** - `packages/shared/src/server/repositories/traces.ts:1026-1113`
+- [x] **getScoresForPostHog()** - `packages/shared/src/server/repositories/scores.ts:1328`
+- [x] **getGenerationsForPosthog()** - `packages/shared/src/server/repositories/observations.ts:1481`
+- [x] **getTracesForBlobStorageExport()** - `packages/shared/src/server/repositories/traces.ts:980-1024`
 
 ### 6. Count and Statistics Queries