mirror of
https://github.com/hanzoai/brain.git
synced 2026-10-30 18:46:38 +00:00
Closed every remaining π‘ / π΅ row in PARITY.md. Each algorithm FortΓ©mi ships now has a real implementation in @hanzo/bot-memory v2.0.0 with matching ports in hanzo-memory (Python), bot-go (Go), and hanzoai/mcp (Rust). Cross-runtime tests: 121 TS + 53 Python + 58 Go + 38 Rust = 270 passing. Landed under packages/memory/ Retrieval & ranking fusion RRF (k=20 BEIR) + RSF (Weaviate v1.24) + adaptive k + adaptive FTS/semantic weights rerank MMR (Carbonell & Goldstein 1998), cosine dedup chunk-aware deduplication script Unicode script detection (Latin/CJK/emoji/Cyrillic/ Arabic/Hebrew/Greek/Devanagari) fts feature-flag matrix, CJK bigrams, emoji trigrams, websearch_to_tsquery parser, SQLite FTS5 MATCH renderer embed embedding registry, asymmetric E5 prefixes, Matryoshka truncation, coarse-dim picker two-stage coarse β fine retrieval via MRL federated cross-archive search with RRF fusion multi-memory X-Hanzo-Memory routing (X-Fortemi-Memory alias) filters composable tag/temporal/collection/scope predicates temporal UUIDv7 floor/ceiling + named ranges spatial Haversine + bbox + in-box (PostGIS fallback) Knowledge & graph graph normalize β SNN β PFNET-β β Louvain maintenance pipeline skos W3C SKOS concept tree (broader/narrower closures, Turtle export) fair Dublin Core XML + JSON-LD doc-types document type registry + auto-detect (filename/MIME/ content sniff) captions WebVTT / SRT / RTTM renderers sprite thumbnail sprite-sheet WebVTT cue builder code-ast language-aware syntactic chunker email RFC 2822 / MIME parser (multipart, qp, base64, HTML strip) exif EXIF parser (camera / time / GPS) Operational events EventBus + SSE renderer + webhook HMAC signer tokenizer BPE-style estimator + truncate-to-budget eval MRR / recall@k / precision@k / NDCG@k / benchmark hardware tier autodetect + per-tier model recommender circuit-breaker closed / open / half-open retry exponential backoff with full jitter range HTTP Range parser + Content-Range formatter tus TUS v1.0.0 (Create / Patch / metadata / checksum) Crypto address wallet-style content-addressable ids (hanzo: / mm:) mmpke01 multi-recipient envelope parser/builder + magic-byte sniff Inference (packages/memory/inference/) index InferenceBackend trait + registry slug provider-qualified routing (provider:model:tag) capabilities capability flags + task-based selector runtime-config hot-swap config (db_override β env β default) mock deterministic backend (for tests) few-shot ICL prompt builder (3-5 examples capped) refinement Self-Refine / ReAct / Reflexion patterns link-types rule + LLM link classifier (11 canonical types) vision VisionBackend trait + Ollama default transcription Whisper-compatible HTTP client diarization pyannote-compatible HTTP client + alignSpeakers / relabel gliner GLiNER zero-shot NER HTTP client Extraction (packages/memory/extract/) index registry + dispatcher text native UTF-8 passthrough email RFC 2822 adapter spreadsheet CSV/TSV pure-JS; xlsx via registerXlsxDecoder archive ZIP central-directory pure-JS reader exif-adapter EXIF for images code tree-sitter-style syntactic chunker pdf PdfSidecar HTTP adapter factory image VisionBackend-backed adapter factory audio TranscriptionBackend-backed adapter factory video keyframe + scene + transcript adapter factory 3d multi-view 3D model adapter factory media-optimize ffmpeg variants (faststart / web-compat / 720p) adapter factory structured JSON-schema structured-extract adapter factory summarizer fast/standard cascade adapter factory sprite-adapter thumbnail sprite-sheet adapter factory Cross-runtime ports hanzo-memory (Python) 53 tests, ports of fusion / rerank / dedup / script / fts / embed / temporal / captions / tokenizer / eval / spatial / range / address / graph / doc-types / circuit / retry / slug / runtime-config / link-types bot-go (Go) 58 tests, same algorithm set ported with standard-library-only primitives hanzoai/mcp::brain (Rust) 38 tests, same algorithm set; sha2 added as dependency (blake3 stand-in) PARITY.md is now 100% green (β in-monorepo + π’ sibling-repo). No π΅ placeholders remain.
36 lines
1.1 KiB
TypeScript
36 lines
1.1 KiB
TypeScript
/** RFC 2822 email adapter. */
|
|
import { parseEmail } from "../email.js";
|
|
import type { ExtractionAdapter } from "./index.js";
|
|
|
|
export const emailAdapter: ExtractionAdapter = {
|
|
strategy: "email",
|
|
async extract({ bytes, filename }) {
|
|
const parsed = parseEmail(bytes);
|
|
const headerLines = [
|
|
parsed.from ? `From: ${parsed.from}` : null,
|
|
parsed.to.length ? `To: ${parsed.to.join(", ")}` : null,
|
|
parsed.subject ? `Subject: ${parsed.subject}` : null,
|
|
parsed.date ? `Date: ${parsed.date}` : null,
|
|
].filter(Boolean) as string[];
|
|
const text = `${headerLines.join("\n")}\n\n${parsed.textBody}`;
|
|
return {
|
|
text,
|
|
metadata: {
|
|
from: parsed.from,
|
|
to: parsed.to,
|
|
cc: parsed.cc,
|
|
bcc: parsed.bcc,
|
|
subject: parsed.subject,
|
|
date: parsed.date,
|
|
messageId: parsed.messageId,
|
|
},
|
|
derived: parsed.attachments.map((a, i) => ({
|
|
filename: a.filename ?? `${filename}-attachment-${i}`,
|
|
mimeType: a.mimeType,
|
|
bytes: a.body,
|
|
role: "email-attachment",
|
|
})),
|
|
};
|
|
},
|
|
};
|