Agent Beacon: reusable memory from coding-agent session history

What Beacon preserves

Agent Beacon is an open-source memory layer for coding agents. Its README describes capturing session history across Claude Code, Cursor, Codex, OpenCode and more than 20 other harnesses.

The goal is to preserve useful debugging paths, corrections and repository conventions after a conversation ends. Beacon normalizes activity into a common trace format, with durable JSONL records that can be inspected independently of its dashboard.

History and reusable knowledge are different stages

The documented workflow moves from captured sessions to evaluating what worked, extracting knowledge, reviewing it and making it available to future agents. That review stage matters: a command appearing in a successful session does not automatically make it the right command for another repository or environment.

The local dashboard and terminal trace browser provide ways to inspect prompts, tool activity, edits and other recorded events. Integration coverage differs by harness, so the README’s support table is more informative than assuming every tool records every field.

Choose the destination deliberately

Beacon describes local-first operation and an interactive setup that preselects Beacon Cloud while offering an explicit Local option. Signing in alone does not enable forwarding; confirming the managed Cloud setup does. The project also distinguishes its local MCP server from hosted MCP access.

Because traces can contain substantial development context, a first evaluation should use a non-sensitive repository and inspect the actual retained events and destination. Verify which capture mode is enabled before extending it to more workspaces.

The useful outcome is a small piece of reviewed knowledge that another session retrieves correctly. More recorded history is not automatically better memory; provenance, relevance and a way to revise stale guidance determine whether the record helps future work.