Beacon

Beacon review: a free, MIT-licensed memory layer that shares what your coding agents learn across Claude Code, Cursor, Codex and 20+ others.

free tiermacOS, Windows, Linux (CLI)

Overview

Beacon is a memory layer for coding agents. It records your sessions in Claude Code, Cursor, Codex, OpenCode and 20+ other harnesses, then turns the fixes, corrections and repo conventions into knowledge the next agent can pull in over MCP or Agent Skills.

The open-source edition is free, MIT-licensed and local-first: logs land as JSONL on your machine, and `beacon traces` browses them in the terminal.

We haven't used it; this page is based on Beacon's site, pricing page and GitHub README.

Best for

Developers who switch between Claude Code, Cursor and Codex and are tired of re-explaining the same repo to each one.

Skip it if

Skip it if you use one agent and its own rules file already covers it, or you don't want full session logs on disk.

Pricing

Open SourceFree

$0 forever; MIT license; self-hosted, no account needed.

  • 20+ agents, including Claude Code, Codex and Cursor
  • Local dashboard for sessions and costs
  • Export to Elastic, S3 or any OTel collector
Beacon CloudFree

$0 forever; managed hosting; 50 findings included.

  • Every agent and machine in one timeline
  • Agents search past sessions over MCP
  • Share a session with a link
Cloud Pro$20/mo

Billed monthly; the page shows a 25% saving for annual billing; unlimited findings.

  • Memories agents save and recall over MCP
  • Skills pulled from your sessions
  • Real-time alerts and priority support
EnterpriseCustom

Sales quote (book a demo).

  • SSO, RBAC and audit logs
  • Your own bucket and region
  • Block or require approval for risky actions

Beacon is free as MIT open source or on the hosted Beacon Cloud plan; Cloud Pro is $20/mo, and Enterprise is custom. Checked Oct 21, 2026 on the official pricing page.

Pros

  • One memory, many agents. It works across Claude Code, Cursor, Codex, OpenCode and 20+ other harnesses.
  • Free and MIT. The open-source edition costs nothing and needs no account.
  • Local-first. Sessions are stored as JSONL on your machine, and you can browse them offline.
  • Full replay. Prompts, tool calls, commands, edits and token use sit in one trace.

Cons

  • Cloud is the default. Interactive setup preselects Beacon Cloud, so you have to choose Local to keep sessions off its servers.
  • It edits your agent setup. The wizard installs Beacon Skills into ~/.agents/skills and ~/.claude/skills unless you turn that off.
  • It logs everything. It captures full prompts, responses, commands and edits, so secrets you paste into an agent end up in its logs.
  • Young project. Its GitHub repo was only created in May 2026.

How it compares

Claude Code's CLAUDE.md and Cursor's rules files store memory for one tool, and you write them by hand. Beacon captures sessions across every agent you use and extracts the lessons for you.

Stick with rules files if you use one agent. Try Beacon if you switch between several and keep re-teaching them.

Quick start

  1. On macOS, run brew trust asymptote-labs/tap, brew tap asymptote-labs/tap and brew install beacon.
  2. Run beacon endpoint install and choose Local unless you want Beacon Cloud.
  3. Use Claude Code, Cursor or Codex as usual.
  4. Run beacon traces to browse your captured sessions.
  5. Turn off Beacon Skills in the wizard if you don't want it editing ~/.claude/skills.

Verdict

Beacon is worth a look if you switch between several coding agents and keep re-teaching each one your repo. Skip it if you use one agent, or don't want full session logs on disk. Choose Local during setup.

Twice a week · Mon & Thu

Two short emails a week. The best new AI tools, and which one to use.

Free. No spam, unsubscribe anytime.