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.
78 lines
2.4 KiB
TypeScript
78 lines
2.4 KiB
TypeScript
/**
|
|
* Audio transcription backend trait. Concrete implementations talk to
|
|
* Whisper / faster-whisper / whisper.cpp / cloud STT APIs.
|
|
*/
|
|
|
|
export interface TranscriptionSegment {
|
|
startSecs: number;
|
|
endSecs: number;
|
|
text: string;
|
|
/** Optional per-word timestamps from word-level alignment. */
|
|
words?: Array<{ word: string; startSecs: number; endSecs: number }>;
|
|
}
|
|
|
|
export interface TranscriptionRequest {
|
|
audio: Uint8Array;
|
|
language?: string;
|
|
/** Optional offset for chunked uploads. */
|
|
startOffsetSecs?: number;
|
|
}
|
|
|
|
export interface TranscriptionResponse {
|
|
segments: TranscriptionSegment[];
|
|
language?: string;
|
|
model?: string;
|
|
}
|
|
|
|
export interface TranscriptionBackend {
|
|
transcribe(req: TranscriptionRequest): Promise<TranscriptionResponse>;
|
|
}
|
|
|
|
// ββ Whisper-compatible HTTP client βββββββββββββββββββββββββββββββββββββ
|
|
|
|
export interface WhisperHttpOpts {
|
|
baseUrl?: string;
|
|
apiKey?: string;
|
|
model?: string;
|
|
fetchImpl?: typeof fetch;
|
|
}
|
|
|
|
export class WhisperHttp implements TranscriptionBackend {
|
|
private baseUrl: string;
|
|
private apiKey?: string;
|
|
private model: string;
|
|
private fetchImpl: typeof fetch;
|
|
|
|
constructor(opts: WhisperHttpOpts = {}) {
|
|
this.baseUrl = opts.baseUrl ?? "http://whisper:8000";
|
|
this.apiKey = opts.apiKey;
|
|
this.model = opts.model ?? "whisper-1";
|
|
this.fetchImpl = opts.fetchImpl ?? globalThis.fetch;
|
|
}
|
|
|
|
async transcribe(req: TranscriptionRequest): Promise<TranscriptionResponse> {
|
|
const form = new FormData();
|
|
form.append("file", new Blob([new Uint8Array(req.audio)]), "audio");
|
|
form.append("model", this.model);
|
|
form.append("response_format", "verbose_json");
|
|
if (req.language) form.append("language", req.language);
|
|
const res = await this.fetchImpl(`${this.baseUrl}/v1/audio/transcriptions`, {
|
|
method: "POST",
|
|
headers: this.apiKey ? { Authorization: `Bearer ${this.apiKey}` } : undefined,
|
|
body: form,
|
|
});
|
|
if (!res.ok) throw new Error(`whisper: ${res.status} ${res.statusText}`);
|
|
const data = (await res.json()) as { segments?: Array<{ start: number; end: number; text: string }>, language?: string };
|
|
const offset = req.startOffsetSecs ?? 0;
|
|
return {
|
|
segments: (data.segments ?? []).map((s) => ({
|
|
startSecs: s.start + offset,
|
|
endSecs: s.end + offset,
|
|
text: s.text,
|
|
})),
|
|
language: data.language,
|
|
model: this.model,
|
|
};
|
|
}
|
|
}
|