MarkdownKeeper v1.0.0

First stable release of MarkdownKeeper — an LLM-accessible markdown documentation database with hybrid semantic search, a durable file watcher, a JSON-RPC HTTP API, and systemd lifecycle management.

Highlights

• Hybrid semantic search — document-vector + chunk-vector + lexical + concept + freshness signals.
• Two-tier embeddings — sentence-transformers (all-MiniLM-L6-v2) with a deterministic SHA-256 hash fallback that needs no ML dependencies.
• Durable watcher — SQLite-persisted event queue, coalescing, idempotent create/modify/move/delete, restart-safe replay.
• JSON-RPC API — /api/v1/query, /api/v1/get_doc, /api/v1/find_concept, /health with token-budgeted, section-filtered delivery.
• Ops — hardened systemd units, full daemon lifecycle, metrics/logging, operations runbook.

Verified KPIs (25-doc fixture corpus, real MiniLM model)

KPI	Target	Measured
precision@5	≥ 0.90	1.000
search latency p95	< 150 ms	0.3 ms
embed 25 documents	< 30 s	13.5 s

Compatibility

CLI commands, machine-oriented JSON keys, /api/v1/* paths and RPC method names, and the initialize_database storage entry point are frozen for the 1.x line. See docs/COMPATIBILITY.md.

Install

pip install markdownkeeper            # hash-fallback embeddings, no ML deps
pip install 'markdownkeeper[embeddings]'   # model-backed semantic search
pip install 'markdownkeeper[embeddings,faiss]'  # + FAISS acceleration

Full notes: CHANGELOG.md.