• Python LLM agent with long-term memory

  shippedPython 3.11+

  A CLI chat agent whose entire long-term memory is one local .sqlrite file. Embeds each turn, hybrid-searches messages + summaries + a structured facts table on every recall, and persists across process restarts. No Postgres, no Redis, no Pinecone — just one file.

  • Vector KNN over past turns via HNSW, plus BM25 keyword recall via fts_match / bm25_score
  • Heuristic fact extraction into a (subject, predicate, object) table — surfaced via plain SQL
  • Zero-config first-run with a hash embedder + offline echo agent; swap in OpenAI / sentence-transformers / Anthropic via CLI flags
  • 31 offline tests; runs end-to-end without an API key
  HNSWVECTOR(384)BM25 / FTSPyO3 SDK
  View on GitHub →Read the README →

• Chat with your notes — Claude Desktop + MCP

  shippedNode.js 20+

  A Node.js CLI that ingests a folder of markdown notes into a SQLRite database, then exposes it to Claude Desktop (or any MCP client) via sqlrite-mcp --read-only. Claude calls bm25_search / vector_search / query directly against your local notes — no cloud sync, no custom RAG pipeline.

  • Markdown → frontmatter-aware chunker → hash or OpenAI embedder → SQLRite documents + chunks tables
  • Hybrid retrieval fuses BM25 and vector cosine in a single SQL ORDER BY (see docs/fts.md)
  • `sqlrite-notes serve` wraps sqlrite-mcp so the Claude Desktop config snippet is one block of JSON
  • Default embedder is fully offline (zero-dep hash bag-of-words); flip to text-embedding-3-small with OPENAI_API_KEY
  • 40 unit + integration tests; works against the prebuilt @joaoh82/sqlrite npm binaries
  HNSWBM25 / FTSMCP servernapi-rs SDK
  View on GitHub →Read the README →

• Local-first journaling — Tauri 2 + Svelte 5

  shippedRust + Svelte 5

  A markdown daily-notes desktop app whose entire data layer is one .sqlrite file you can copy between machines, back up with rsync, or open in the SQLRite REPL. Phase 8 BM25 full-text search with hit highlighting, click-to-filter tags, and an 'ask my journal' panel that turns natural language into read-only SQL against your own entries.

  
  • Tauri 2 backend owns an Arc<Mutex<Connection>> in tauri::State — commands serialise through one engine handle, no torn writes
  • BM25 full-text search over entry content with token-boundary-aware <mark> highlighting on the Svelte side
  • 'Ask my journal' panel calls Connection::ask, validates the returned SQL is SELECT/WITH-only, and shows you both the SQL and the rows — API key never crosses into the webview
  • Export options: copy the .sqlrite file as-is, or dump every entry as YAML-frontmatter markdown into a folder
  • Locked-down Tauri capabilities (core window + dialog only — no fs / shell / http)
  Connection APIBM25 / FTSaskTauri 2
  View on GitHub →Read the README →

• Browser SQL playground — WebAssembly

  shippedWebAssembly

  The full SQLRite engine compiled to WebAssembly, running entirely in a browser tab — no server, no install, no account. Type SQL, hit Run, see results. Load a Pokémon / Northwind / movies-with-embeddings dataset in one click, then poke at JOINs, GROUP BY, and HNSW cosine vector search right on the page.

  • Same sqlrite-engine crate as every other SDK, built for wasm32 via @joaoh82/sqlrite-wasm (~750 KB gzipped, fetched once)
  • CodeMirror 6 editor with SQL highlighting + Cmd/Ctrl+Enter; results grid with column types, NULL highlighting, and CSV export
  • Vector-search demo dataset: 12 films with 4-dim embeddings + an HNSW index, ranked by vec_distance_cosine in the browser
  • Session persists to OPFS (localStorage fallback) via SQL-script replay; share a query by URL hash; download / upload .sql
  • Hosted on this site — open it live, no clone required
  WASM SDKHNSWCodeMirror 6OPFS
  ▸ Try it liveView on GitHub →Read the README →

• Edge / IoT event collector — Go

  shippedGo 1.21+

  A Go service that buffers telemetry from many concurrent HTTP producers into a local .sqlrite file, while a background uploader goroutine drains it to a remote sink — both writing the same database at once via BEGIN CONCURRENT (Phase 11 MVCC). The .sqlrite file is the durable buffer between an unreliable network and the upstream system: it survives reboots and stays queryable with SQL on-device.

  • HTTP write path and uploader each hold their own BEGIN CONCURRENT transaction against one engine via the Go driver's sibling-handle registry — multi-writer concurrency a single-writer DB can't express
  • Row-level conflict detection + the canonical retry-on-ErrBusy loop; zero dropped events under 32 concurrent producers (verified by the load generator)
  • Honest, measured throughput: on v0 BEGIN CONCURRENT is ~0.9× a single mutex (the win is capability + correctness, not raw speed) — the README ships all three measurement tables and explains why
  • Documents the v0 engine sharp edges it designs around: no param binding, 4 KiB MVCC commit cap, app-assigned ids under MVCC, reopen detection without a queryable catalog
  • Backpressure (503) + /healthz + /stats; Dockerfile + compose with an optional read-only sqlrite-mcp sidecar; Linux + macOS CI against the in-repo engine
  Go SDK (cgo)BEGIN CONCURRENTMVCCsqlrite-ffi
  View on GitHub →Read the README →

All five umbrella examples have shipped. See /docs for the engine reference.