Records coding-agent sessions locally and turns the useful lessons into reusable skills.
What it does
On this page
Beacon watches what your coding agents do on a machine, stores each session as a series of events, and pulls repeatable lessons out of them. The project covers Claude Code, Cursor, Codex, OpenCode and a longer list of harnesses.
Who it is for
Developers who run more than one coding agent and want the next session to start with the corrections from the last one.
How it works
A CLI endpoint runs locally and records prompts, commands, diffs and approvals. Sessions land in a local dashboard first. From there you can read them, forward them to your own Elastic, S3 or OpenTelemetry stack, or let Beacon Cloud keep one timeline across machines. The paid tier adds findings, memory the agent can query over MCP, and skills generated from your own sessions. The CLI and threat rules are open source under the MIT licence.
Pricing
Open source and Beacon Cloud cost nothing. Cloud Pro is $20 a month, or $15 a month on annual billing. Team adds a shared workspace at $30 per seat a month plus a $10 base fee, capped at 20 seats. Enterprise is quoted separately.
Notable limits
Memory over MCP and unlimited findings belong to Pro; a free cloud account includes 50 findings.
Source: https://beacon.sh/