Files
zeekayandGitHub 1e38b6fc1d feat(rag): consolidate file-scoped RAG into ai; retire chat-rag-api (#35) (#59)
* feat(rag/split): recursive character splitter for rag-api chunk parity

The RAG surface (/v1/rag/*, object.Rag*) previously relied on the retired
standalone rag-api's RecursiveCharacterTextSplitter for chunking. Fold that
into the ONE splitting home: RecursiveSplitProvider replicates LangChain's
separator cascade (paragraph -> line -> word -> char), keeping every chunk at
or below ChunkSize with ChunkOverlap-sized overlap, rune-measured for Unicode
parity with Python len(). Zero-value construction yields the retired service's
CHUNK_SIZE=1500 / CHUNK_OVERLAP=100 defaults, and GetSplitProvider("Recursive")
wires it into the factory so retrieval recall matches the service it replaces.

Hermetic unit tests (recursive_test.go) cover parity defaults, overlap
clamping, the chunk-size bound, single-chunk short text, overlap duplication,
and factory wiring — all green with `go test -run Recursive`. The pre-existing
TestSplit/2/3 integration harnesses in provider_test.go stay env-dependent and
are unrelated to this change.

* feat(rag): consolidate file-scoped RAG into ai; retire chat-rag-api (#35)

File-scoped RAG (LibreChat/hanzo.chat's per-file_id contract) folds into the
ONE unified RAG index rather than a parallel store: file_id becomes a
filterable dimension on {owner}-{store}-docs (Meili filterable attr + Qdrant
payload), so uploaded-file retrieval is a filter over the same Hanzo Search +
Hanzo Vector pipeline that doc/crawl/github/s3 ingest already use.

- object/rag.go: RagEmbedFile / RagQuery / DeleteRagFile / RagFileContext,
  reusing IndexDocuments/SearchDocuments, txt parsers, and the Recursive
  splitter (chunk parity 1500/100). Uploaded files default to the rag-files
  store. Re-embed replaces a file_id's chunks (idempotent).
- object/search_docs.go: thread file_id end-to-end (DocIndex.FileID,
  DocSearchRequest.FileIDs, buildMeiliFilter with injection-safe quoting,
  Qdrant should-clause, filtered delete seam, file-context fetch). Additive,
  backward-compatible.
- controllers/rag.go: canonical /v1/rag/{embed,query,query-multiple,delete,context}.
- controllers/rag_librechat.go: drop-in for the retired chat-rag-api contract
  (POST /v1/embed multipart, /v1/query, /v1/query_multiple, DELETE /v1/documents,
  GET /v1/documents/:file_id/context) returning the LangChain tuple shape
  hanzo.chat parses. Point RAG_API_URL at api.hanzo.ai/v1 -> no chat-repo change.
- tests: hermetic sink-swap coverage of the full embed->query->delete lifecycle,
  file-scope isolation, chunking, and the compat shape adapters.

Auth/billing reuse the doc-RAG path; owner is the authenticated principal
(tenant isolation), never client-supplied.
2026-07-02 21:17:18 -07:00

309 lines
9.6 KiB
Go

// Copyright 2023-2025 Hanzo AI Inc. All Rights Reserved.
//
// Licensed under the Apache License, Version 2.0 (the "License");
// you may not use this file except in compliance with the License.
// You may obtain a copy of the License at
//
// http://www.apache.org/licenses/LICENSE-2.0
//
// Unless required by applicable law or agreed to in writing, software
// distributed under the License is distributed on an "AS IS" BASIS,
// WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
// See the License for the specific language governing permissions and
// limitations under the License.
package controllers
// LibreChat-compat RAG surface — the drop-in replacement for the retired
// standalone chat-rag-api (danny-avila/rag_api fork). hanzo.chat's RAG client
// (api/server/services/Files/VectorDB/crud.js + createContextHandlers.js) calls
// a FIXED contract at RAG_API_URL:
//
// POST {base}/embed multipart(file_id,file[,entity_id])
// -> {status, known_type, file_id, filename}
// POST {base}/query {file_id,query,k} -> [[{page_content,metadata},score],...]
// POST {base}/query_multiple {file_ids,query,k} -> same tuple shape
// DELETE {base}/documents [file_id,...]
// GET {base}/documents/{file_id}/context -> [{page_content,metadata},...]
//
// Pointing RAG_API_URL at https://api.hanzo.ai/v1 makes these resolve here with
// ZERO chat-repo change. The handlers are a thin HTTP projection over the ONE
// canonical RAG logic (object.Rag*); they hold no retrieval logic of their own.
// The native /v1/rag/* surface (controllers/rag.go) shares that same logic.
import (
"encoding/json"
"io"
"os"
"path/filepath"
"strings"
"github.com/hanzoai/ai/object"
"github.com/hanzoai/ai/txt"
)
// lcDocument mirrors LangChain's Document JSON ({page_content, metadata}) that
// the retired rag-api returned and hanzo.chat's client parses (item[0].page_content).
type lcDocument struct {
PageContent string `json:"page_content"`
Metadata map[string]interface{} `json:"metadata"`
}
// lcResultToDocument converts a canonical DocSearchResult into the LangChain
// Document shape, preserving file_id/title/url in metadata as the old rag-api did.
func lcResultToDocument(r object.DocSearchResult) lcDocument {
return lcDocument{
PageContent: r.Content,
Metadata: map[string]interface{}{
"file_id": r.FileID,
"title": r.Title,
"url": r.URL,
"id": r.ID,
},
}
}
// lcTuples renders results as [[document, score], ...]. The retired rag-api
// returned (Document, distance) tuples; hanzo.chat only reads item[0], so the
// score slot carries the rank position (lower = better) as a stable placeholder
// — the client does not use it, and the native /v1/rag/query exposes real scores.
func lcTuples(results []object.DocSearchResult) []interface{} {
out := make([]interface{}, 0, len(results))
for i, r := range results {
out = append(out, []interface{}{lcResultToDocument(r), i})
}
return out
}
// isKnownType reports whether the filename's extension is a parser-supported
// type. hanzo.chat treats known_type=false as "filetype not supported".
func isKnownType(filename string) bool {
ext := strings.ToLower(filepath.Ext(filename))
if ext == "" {
return false
}
for _, t := range txt.GetSupportedFileTypes() {
if t == ext {
return true
}
}
// Source-code / plaintext-ish files are parsed as plain text.
return true
}
// RagEmbedMultipart handles POST /v1/embed — the multipart upload hanzo.chat's
// uploadVectors() sends. It saves the upload to a temp file, then reuses
// object.RagEmbedFile (same parse+chunk+index as the native surface).
//
// @Title RagEmbedMultipart
// @Tag RAG API (LibreChat-compat)
// @Description Multipart file embed (drop-in for the retired rag-api POST /embed).
// @router /embed [post]
func (c *ApiController) RagEmbedMultipart() {
auth := c.requireIndexAuth()
if auth == nil {
return
}
fileID := c.GetString("file_id")
if fileID == "" {
c.ragCompatError("file_id is required", "")
return
}
entityID := c.GetString("entity_id") // reserved for shared-resource scoping
f, header, err := c.GetFile("file")
if err != nil {
c.ragCompatError("file is required: "+err.Error(), fileID)
return
}
defer f.Close()
filename := header.Filename
if !isKnownType(filename) {
// Match the retired rag-api: report the unsupported type, don't 500.
c.Data["json"] = map[string]interface{}{
"status": false, "known_type": false, "file_id": fileID, "filename": filename,
}
c.ServeJSON()
return
}
tmpPath, err := saveUploadToTemp(f, filename)
if err != nil {
c.ragCompatError("failed to buffer upload: "+err.Error(), fileID)
return
}
defer os.Remove(tmpPath)
// URL as a local path -> object.RagEmbedFile parses via txt.GetParsedTextFromUrl
// (which treats a non-http url as a path), the SAME parser doc-ingest uses.
req := &object.RagEmbedRequest{
FileID: fileID,
Filename: filename,
URL: tmpPath,
Tag: entityID,
}
result, err := object.RagEmbedFile(auth.Owner, req, c.GetAcceptLanguage())
if err != nil {
recordSearchUsage(auth, "index-docs", "rag-embed", "error", 0, c.Ctx.Request.RemoteAddr)
c.ragCompatError(err.Error(), fileID)
return
}
recordSearchUsage(auth, "index-docs", "rag-embed", "success", result.Chunks, c.Ctx.Request.RemoteAddr)
c.Data["json"] = map[string]interface{}{
"status": true,
"known_type": true,
"file_id": fileID,
"filename": filename,
"chunks": result.Chunks,
}
c.ServeJSON()
}
// RagQueryCompat handles POST /v1/query — {file_id,query,k}. Returns LangChain
// (document,score) tuples.
//
// @Title RagQueryCompat
// @Tag RAG API (LibreChat-compat)
// @Description File-scoped query (drop-in for the retired rag-api POST /query).
// @router /query [post]
func (c *ApiController) RagQueryCompat() {
c.ragQueryCompat()
}
// RagQueryMultipleCompat handles POST /v1/query_multiple — {file_ids,query,k}.
//
// @Title RagQueryMultipleCompat
// @Tag RAG API (LibreChat-compat)
// @Description Multi-file query (drop-in for the retired rag-api POST /query_multiple).
// @router /query_multiple [post]
func (c *ApiController) RagQueryMultipleCompat() {
c.ragQueryCompat()
}
// ragQueryCompat is the shared read path for /v1/query and /v1/query_multiple.
// The canonical RagQueryRequest already carries both file_id and file_ids.
func (c *ApiController) ragQueryCompat() {
auth := c.resolveSearchAuth()
if auth == nil {
return
}
var req object.RagQueryRequest
if err := json.Unmarshal(c.Ctx.Input.RequestBody, &req); err != nil {
c.ResponseError(err.Error())
return
}
results, err := object.RagQuery(auth.Owner, &req, c.GetAcceptLanguage())
if err != nil {
recordSearchUsage(auth, "search-query", "rag", "error", 0, c.Ctx.Request.RemoteAddr)
c.ResponseError(err.Error())
return
}
recordSearchUsage(auth, "search-query", "rag", "success", len(results), c.Ctx.Request.RemoteAddr)
c.Data["json"] = lcTuples(results)
c.ServeJSON()
}
// RagDeleteDocuments handles DELETE /v1/documents — a JSON array of file_ids.
//
// @Title RagDeleteDocuments
// @Tag RAG API (LibreChat-compat)
// @Description Delete files by id (drop-in for the retired rag-api DELETE /documents).
// @router /documents [delete]
func (c *ApiController) RagDeleteDocuments() {
auth := c.requireIndexAuth()
if auth == nil {
return
}
var fileIDs []string
if err := json.Unmarshal(c.Ctx.Input.RequestBody, &fileIDs); err != nil {
c.ResponseError(err.Error())
return
}
if len(fileIDs) == 0 {
c.ResponseError("expected a non-empty array of file_ids")
return
}
lang := c.GetAcceptLanguage()
deleted := 0
for _, id := range fileIDs {
if id == "" {
continue
}
if err := object.DeleteRagFile(auth.Owner, "", id, lang); err != nil {
c.ResponseError(err.Error())
return
}
deleted++
}
c.Data["json"] = map[string]interface{}{
"message": "Documents deleted successfully",
"deleted": deleted,
}
c.ServeJSON()
}
// RagDocumentContext handles GET /v1/documents/:file_id/context — every chunk of
// a file, as LangChain Documents (used when RAG_USE_FULL_CONTEXT is on).
//
// @Title RagDocumentContext
// @Tag RAG API (LibreChat-compat)
// @Description Full file context (drop-in for rag-api GET /documents/{id}/context).
// @router /documents/:file_id/context [get]
func (c *ApiController) RagDocumentContext() {
auth := c.resolveSearchAuth()
if auth == nil {
return
}
fileID := c.Ctx.Input.Param(":file_id")
if fileID == "" {
c.ResponseError("file_id must not be empty")
return
}
results, err := object.RagFileContext(auth.Owner, "", fileID)
if err != nil {
c.ResponseError(err.Error())
return
}
docs := make([]lcDocument, 0, len(results))
for _, r := range results {
docs = append(docs, lcResultToDocument(r))
}
c.Data["json"] = docs
c.ServeJSON()
}
// ragCompatError emits the retired rag-api's failure shape ({status:false,...})
// so hanzo.chat's uploadVectors() surfaces a clean "File embedding failed".
func (c *ApiController) ragCompatError(msg, fileID string) {
c.Data["json"] = map[string]interface{}{
"status": false, "known_type": true, "file_id": fileID, "detail": msg,
}
c.ServeJSON()
}
// saveUploadToTemp streams an uploaded file to a temp path, preserving the
// extension so the parser can dispatch on it.
func saveUploadToTemp(f io.Reader, filename string) (string, error) {
ext := filepath.Ext(filename)
tmp, err := os.CreateTemp("", "rag-upload-*"+ext)
if err != nil {
return "", err
}
defer tmp.Close()
if _, err := io.Copy(tmp, f); err != nil {
os.Remove(tmp.Name())
return "", err
}
return tmp.Name(), nil
}