feat(brain): hybrid BM25 + pgvector retrieval (opt-in)
All checks were successful
CI / Lint / Test / Vet (push) Successful in 15s
CI / Mirror to GitHub (push) Successful in 3s

Wires nomic-embed-text (iguana ollama) + pgvector on the shared
postgres18 into brain_query / brain_answer via Reciprocal Rank Fusion.
Pure BM25 stays the default; setting BRAIN_PG_DSN and BRAIN_EMBED_URL
together opts in. Setting one without the other is misconfiguration →
exit 1.

New packages:

- internal/embed
  Client.Embed(ctx, text) → []float32 via POST {URL}/api/embed.
  Defaults to nomic-embed-text:latest (768 dim). nil-on-empty-URL so
  callers gate on a single nil check.

- internal/vectorstore
  PGStore wraps a pgxpool against postgres18. Init creates
  brain_embeddings(path PK, vector(768), updated_at) + HNSW cosine
  index idempotently. Upsert / Delete / Search / KnownPaths.
  Sync(brainDir, store, embedder) diffs brain/wiki/ against the store
  and upserts new files / deletes removed ones; StartSync runs it on
  a ticker (default 300s). Integration tests gated by BRAIN_PG_TEST_DSN.

- scripts/brain-embeddings-init.sql
  One-time DBA setup: brain DB, brain_app role, vector extension,
  GRANTs. Idempotent.

Search layer:

- search.QueryOptions gains Vector + Embedder fields.
- QueryContext is the cancellable variant; Query stays for callers.
- When both are set, BM25 (top-N) and pgvector (top-4N) candidates
  merge via Reciprocal Rank Fusion (k=60, Cormack et al. 2009 — no
  tuning knob, robust to scale differences between rankers).
- Vector-only hits are hydrated from disk so callers see uniform
  Result records (path, title, excerpt, wing, hall, score).
- Wing/hall filters still apply to vector candidates via path-prefix.
- On embedder/vector errors the search falls back to BM25 — embedding
  outage degrades quality but doesn't take the brain offline.

MCP wiring:

- mcp.Server.WithHybridRetrieval(v, e) opt-in setter, same shape as
  WithReranker.
- brainQuery and brainAnswer pass the wired vector/embedder through
  to search.QueryContext.

REST:

- POST /backfill-embeddings drives Sync synchronously. Returns
  {added, deleted, errors[]}. 503 when feature is unconfigured.

cmd/server/main.go:

- BRAIN_PG_DSN + BRAIN_EMBED_URL together enable hybrid; one alone
  → exit 1.
- vectorAdapter bridges *PGStore (returns []Hit) to
  search.VectorSearcher (which takes []VectorHit) without either
  package importing the other.
- BRAIN_EMBED_SYNC_INTERVAL (default 300s) controls the background
  Sync ticker.

Backend pivot from Qdrant to pgvector recorded in DECISIONS.md
2026-05-18 (supersedes 2026-04-08): postgres18 already runs in
databases/ ns, Qdrant was never deployed, one engine beats two.

Dependency: github.com/jackc/pgx/v5 — modern, native pgvector via
parametric vector literals.

Tests:
- embed.Client: empty-URL nil, request shape, dimension, upstream
  error propagation, empty-text rejection.
- vectorstore.PGStore: dimension validation (unit); upsert/search/
  KnownPaths (integration, BRAIN_PG_TEST_DSN-gated).
- vectorstore.Sync: adds new files, skips known, deletes
  disappeared, skips _index.md, no-op when nil, collects embedder
  errors.
- search.Query: hybrid promotes vector-only hits via RRF; falls
  back to BM25 on embedder error.

Closes hyperguild#8.
This commit is contained in:
Mathias
2026-05-18 23:11:25 +02:00
parent a56a4db963
commit 57462b52ff
16 changed files with 1068 additions and 14 deletions

View File

@@ -15,12 +15,32 @@ import (
"github.com/mathiasbq/hyperguild/ingestion/internal/auth"
"github.com/mathiasbq/hyperguild/ingestion/internal/llm"
"github.com/mathiasbq/hyperguild/ingestion/internal/mcp"
"github.com/mathiasbq/hyperguild/ingestion/internal/embed"
"github.com/mathiasbq/hyperguild/ingestion/internal/oauth"
"github.com/mathiasbq/hyperguild/ingestion/internal/reranker"
"github.com/mathiasbq/hyperguild/ingestion/internal/pipeline"
"github.com/mathiasbq/hyperguild/ingestion/internal/reranker"
"github.com/mathiasbq/hyperguild/ingestion/internal/search"
"github.com/mathiasbq/hyperguild/ingestion/internal/vectorstore"
"github.com/mathiasbq/hyperguild/ingestion/internal/watcher"
)
// vectorAdapter bridges *vectorstore.PGStore (returns []vectorstore.Hit)
// to the search.VectorSearcher interface (which uses []search.VectorHit).
// Kept here, not in either package, so neither has to import the other.
type vectorAdapter struct{ s *vectorstore.PGStore }
func (a vectorAdapter) Search(ctx context.Context, q []float32, limit int) ([]search.VectorHit, error) {
hits, err := a.s.Search(ctx, q, limit)
if err != nil {
return nil, err
}
out := make([]search.VectorHit, len(hits))
for i, h := range hits {
out[i] = search.VectorHit{Path: h.Path, Distance: h.Distance}
}
return out, nil
}
func envOr(key, fallback string) string {
if v := os.Getenv(key); v != "" {
return v
@@ -84,6 +104,37 @@ func main() {
logger.Info("brain reranker configured", "url", rerankURL, "model", rerankModel)
}
// Hybrid retrieval (pgvector + nomic-embed-text). Both env vars must
// be set together for the path to wire on; otherwise BM25-only.
var vectorStore *vectorstore.PGStore
pgDSN := os.Getenv("BRAIN_PG_DSN")
embedURL := os.Getenv("BRAIN_EMBED_URL")
switch {
case pgDSN != "" && embedURL != "":
embedModel := envOr("BRAIN_EMBED_MODEL", "nomic-embed-text:latest")
store, err := vectorstore.New(context.Background(), pgDSN)
if err != nil {
logger.Error("vector store init", "err", err)
os.Exit(1)
}
if err := store.Init(context.Background()); err != nil {
logger.Error("vector store migrate", "err", err)
os.Exit(1)
}
vectorStore = store
embedder := embed.New(embedURL, embedModel)
mcpSrv = mcpSrv.WithHybridRetrieval(vectorAdapter{s: store}, embedder)
h.WithEmbedSync(store, embedder)
logger.Info("brain hybrid retrieval enabled",
"pg", pgDSN[:strings.IndexByte(pgDSN+"@", '@')], // crude redaction
"embed_url", embedURL, "embed_model", embedModel)
case pgDSN == "" && embedURL == "":
// disabled — fine
default:
logger.Error("BRAIN_PG_DSN and BRAIN_EMBED_URL must be set together")
os.Exit(1)
}
mcpToken := os.Getenv("BRAIN_MCP_TOKEN")
if mcpToken == "" {
logger.Error("BRAIN_MCP_TOKEN not set")
@@ -98,6 +149,14 @@ func main() {
Pipeline: pipelineCfg,
})
}
if vectorStore != nil {
embedSyncInterval := envInt("BRAIN_EMBED_SYNC_INTERVAL", 300)
vectorstore.StartSync(ctx, brainDir, vectorStore,
embed.New(os.Getenv("BRAIN_EMBED_URL"),
envOr("BRAIN_EMBED_MODEL", "nomic-embed-text:latest")),
time.Duration(embedSyncInterval)*time.Second)
logger.Info("embed sync started", "interval_s", embedSyncInterval)
}
mux := http.NewServeMux()
mux.HandleFunc("POST /query", h.Query)
@@ -107,6 +166,7 @@ func main() {
mux.HandleFunc("POST /ingest-path", h.IngestPath)
mux.HandleFunc("POST /ingest-raw", h.IngestRaw)
mux.HandleFunc("POST /backfill-refs", h.BackfillRefs)
mux.HandleFunc("POST /backfill-embeddings", h.BackfillEmbeddings)
mux.HandleFunc("GET /pass-rate", h.PassRate)
var jwtValidator *auth.Validator
if dexURL := os.Getenv("DEX_ISSUER_URL"); dexURL != "" {