Lector

Filesystem & code-intelligence service: a platform-neutral capability core plus a supervised daemon, with host adapters (starting with Pi) as thin translation layers on top.

Filesystem operations and code intelligence are not two capabilities that happen to share a daemon — they operate on the same object, a file's content at a point in time, just through different lenses: raw text for reads and hash-guarded edits, an AST/symbol view for code queries. Lector caches both lenses under one shared, content-addressed key, so satisfying one lens's request can warm the other's for free.

Packages

• packages/lector — the capability core and daemon: domain, ports, adapters (local filesystem, typescript-language-server, tree-sitter, SQLite content cache), and the CLI/systemd service.
• packages/pi-lector — the Pi host adapter: overrides read/write/edit and adds symbol, search, Git, repository, and verified package-source tools, all routed through a running Lector daemon.

A future alef-lector (the Alef host adapter) would join this workspace the same way, depending on packages/lector without duplicating any of its domain behavior.

Architecture

Host adapter (packages/pi-lector, ...)
      ↓
authenticated loopback client
      ↓
Lector daemon (auth, dispatch, cache/index generations)
      ↓
domain: rawRead / exactEdit / findWorkspaceSymbols
      ↓
ports: WorkspacePort, SymbolIndexPort, ContentCachePort
      ↓
adapters: local filesystem, typescript-language-server, tree-sitter (WASM), SQLite

A supervised Bun daemon built on @danypops/daemon-kit: one process owns the database and any subprocesses (language servers); everything else is an authenticated client.

Development

This is a Bun workspace (packages/*) — one bun install at the repo root links pi-lector's dependency on lector locally, no publish needed to develop both together.

bun install     # from the repo root
bun test        # runs every package's tests
bun run typecheck

See each package's own README for its CLI, service, and API surface.