engram
The engine that stores and recalls — polytemporal schema, salience gating, retention profiles.
PROJECT
PAI — agent interface
The polytemporal memory interface your agent actually calls.
pip install "git+https://github.com/ShingWong/positronic-agent-interface.git"CopyThe memory engine decides what to keep. The agent decides what a memory means right now. PAI is the stable seam between them, so opencode and Claude Code stay thin: they spawn a CLI verb and read JSON, while all policy, config and federation live in one place.
The seam, not the engine
PAI is the thin Python layer between your agent and engram: `.positronic/config.json` + `.positronic/brains/{name}/memory.db`.
One contract, many hosts
opencode slashes, agentic tools, and the CLI all dispatch to the same `run(...) -> dict` — JSON-serializable, ready for a tool call.
The frontal lobe is the agent
PAI preserves polytemporal structure and hands it over; it never pre-judges which memory answers. That is the agent's job.
Recall digests
`recall "
Distilled recall
`--consolidation only|first` returns conclusions, not chatter; the consolidation channel is kind-scoped so live chatter can't drown a match.
Full dossiers
`ask "
Strategic direction: ported to WASM + sqlite-wasm + OPFS for shared core across hosts (Thunderbird, Outlook, Personabot); the Python server stays for power users.
Part of the ecosystem
Built alongside
Positronic is a small set of pieces that fit together — an engine, a seam, plugins and a network layer. Here is what sits around this project, and where each piece lives.
opencode plugin
The plugin that wires it into opencode with live auto-ingest on every turn.
claude-code-plugin
The same lifecycle for Claude Code, driven by SessionStart, Stop and PreCompact hooks.
serve
Expose a brain over HTTP, or federate several hosts into one logical memory.
logschema
Measures what your logs actually contain and turns it into typed, retention-aware events — the ingestion side of a brain.