positronic

PROJECT

Positronic for opencode

Federated, polytemporal, tensor-grounded memory that outlives the session.

npm i -g @positronic-ai/opencode-pluginCopy

This is the fastest way to try positronic. Install it, keep working, and every assistant turn is written to a local brain without any manual remember calls or cloud upload. Recall happens before the agent resumes, so context carries across sessions, weeks and entire projects. Nothing is sent anywhere: the brain is a SQLite file sitting next to your project, and you can open it with any tool that reads SQLite. Run positronic stats to see how much has accumulated, positronic recall to query it directly, or stay inside your editor and let the plugin surface what matters. Because retention is a curve rather than a cutoff, older episodes stay reachable instead of expiring on a timer, and because every record is an ordinary row, you can audit exactly what your agent chose to remember.

Auto-ingest

Auto-ingest

Every assistant turn is saved to a local SQLite brain automatically. No cloud. No manual remember calls. No opaque summarizers.

Federated brains

Federated brains

One brain per concern (kairos, mail, research). Query one, or fuse across all with RRF.

Retention curves

Retention curves

balanced / long_term / archival / short_term — survival curves, not TTL. balanced is the everyday default.

Two-tier recall

Two-tier recall

Tier 1 lexical (FTS5) works anywhere; Tier 3 adds BGE-M3 semantic on your GPU.

Local-first

Local-first

One file per brain, offline, reproducible. Live by default — chat.message auto-ingests.

Verified

Verified

Evaluated on LongMemEval, RULER, and the E7 synthetic benchmark.

Legacy unscoped package positronic-opencode-plugin@0.1.0-beta.2 remains published and in use — it is the old beta.

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.

engram

The engine that stores and recalls — polytemporal schema, salience gating, retention profiles.

agent-interface

The CLI and code contract your agent calls to init, ingest, recall and ask.

claude-code-plugin

The same lifecycle for Claude Code, driven by SessionStart, Stop and PreCompact hooks.

research

Papers, architecture notes and the benchmark harness behind every figure quoted here.