Beacon
Beacon is a self-improving, cross-harness memory layer for coding agents. It turns sessions across Claude Code, Codex, Cursor, OpenCode, and 20+ other harnesses into a compounding knowledge layer, so each agent starts smarter.
Visit Asymptote-Labs/agent-beaconOverview
Beacon is a self-improving memory layer for AI coding agents. It captures agent sessions across multiple coding harnesses and turns them into a compounding knowledge layer.
Key Features
- Captures agent sessions across Claude Code, Codex, Cursor, OpenCode, and 20+ other harnesses.
- Turns agent session history into a compounding knowledge layer.
- Uses Jev to identify valuable lessons buried in agent sessions and turn them into reusable skills.
- Helps each agent start smarter than the last by carrying lessons across sessions.
- Works with many popular coding agents and harnesses.
Use Cases
- Carry lessons from previous coding sessions into new agent sessions.
- Reuse high-signal workflows, corrections, and debugging patterns as skills across agents.
- Work across multiple coding harnesses such as Claude Code, Codex, Cursor, and OpenCode.
Getting Started
- Install the Beacon CLI with the provided install command.
- Sign up for a free Beacon account; no credit card is required.
Before You Adopt
- License: MIT. Review its terms before using, modifying, or distributing the project.
- Beacon depends on Jev to evaluate agent sessions and identify valuable lessons for reuse.
