# Beacon

> Open-source, cross-harness memory layer that captures coding agent sessions and turns them into reusable skills for future agents.

Beacon is an open-source memory layer for AI coding agents from Asymptote Labs. It captures session history across Claude Code, Cursor, Codex, OpenCode, and 20+ other harnesses, then turns useful workflows, corrections, and debugging patterns into reusable knowledge for future agents. The project is MIT-licensed, and the latest release listed on GitHub is v1.3.45.

## What It Is

Beacon sits across the agent harness layer rather than inside any single tool. It records prompts, responses, tool calls, commands, file activity, approvals, MCP interactions and token usage, and normalizes them into a common OpenTelemetry-based event model. The result is one session history that can be inspected, searched and learned from, instead of a separate proprietary history per coding tool.

## How the Memory Loop Works

The documented loop is: run agents, capture session history, evaluate what worked, extract useful knowledge, review and approve it, then reuse it across future agents. The website describes a "dreaming" process that reviews successful sessions, identifies the correction that mattered, and turns it into a concrete skill, for example a skill for debugging failed webhook deliveries. Once a skill is reviewed, future agents can retrieve it through MCP or Agent Skills.

## Local-First Setup and Inspection

Beacon installs through Homebrew on macOS, a package installer on Linux (.deb or .rpm), and an MSI on Windows. Telemetry is written to local JSONL by default. `beacon traces` opens a terminal browser for traces, event timelines and token usage, and `beacon endpoint dashboard` opens a local read-only web dashboard. Dashboard sessions can show custom "lenses", sandboxed single-file views such as file diffs or token breakdowns. Interactive setup can instead connect the machine to Beacon Cloud, and `beacon endpoint disconnect` reverses that.

## Coverage and Destinations

The README lists support for local agents, browser chat (Claude.ai and ChatGPT via extension), cloud agents, CI jobs, and SDK instrumentation for Anthropic, OpenAI, the Claude Agent SDK and the Vercel AI SDK. Normalized session history can be forwarded to destinations including Splunk, Datadog, Elastic, Microsoft Sentinel, CrowdStrike Falcon LogScale, Sumo Logic, Wazuh, AWS S3, GCS and CloudWatch.

## Features
- Cross-harness session history capture across 20+ coding agents
- Reusable skills generated from workflows, corrections and debugging patterns
- Shared memory retrievable through MCP or Agent Skills
- Exact session replay of prompts, tools, commands, edits, approvals and tokens
- Local-first JSONL storage
- Local terminal and web dashboard with custom lenses
- Forwarding to SIEM, log aggregation and object storage destinations
- Browser chat, cloud agent, CI and SDK instrumentation

## Integrations
Claude Code, Cursor, Codex, OpenCode, Cline, GitHub Copilot CLI, Gemini CLI, VS Code, Devin, OpenHands, Splunk, Datadog, Elastic, Microsoft Sentinel, CrowdStrike Falcon LogScale, Sumo Logic, Wazuh, AWS S3, Google Cloud Storage, AWS CloudWatch

## Platforms
WINDOWS, MACOS, LINUX, WEB, API, BROWSER_EXTENSION, VSC_EXTENSION, DEVELOPER_SDK, CLI

## Pricing
Open Source, Free tier available

## Version
v1.3.45

## Links
- Website: https://beacon.sh
- Documentation: https://docs.beacon.sh
- Repository: https://github.com/Asymptote-Labs/agent-beacon
- EveryDev.ai: https://www.everydev.ai/tools/beacon