Description
You finally debugged that gotcha in Claude Code this morning, and this afternoon Cursor needs the whole story again. A repo convention you corrected your agent on three times is forgotten the moment you open a new session. Every agent learns on its own, and when the session closes, the experience goes with it.
Beacon is an open-source, local-first memory layer for coding agents. It gathers full session history from Claude Code, Cursor, Codex, OpenCode, Cline and 20+ other harnesses, then turns the useful workflows, your corrections, debugging patterns and repo conventions into knowledge that future agents can reuse. A problem solved by one agent shouldn't have to be learned from scratch by another.
Reviewed knowledge can be retrieved by later agents through MCP or Agent Skills, and you can replay any session exactly, with prompts, responses, tools, commands, edits, approvals and token use in one trace. Data is stored as durable JSONL, you choose the destinations, and no harness has you locked in.
Cross-harness history: sessions from Claude Code, Cursor, Codex, OpenCode, Cline and 20+ other agents in one place.
Knowledge that compounds: workflows, corrections, debugging patterns and repo conventions that outlive a single session.
Shared agent memory: reviewed knowledge that later agents retrieve through MCP or Agent Skills.
Exact session replay: prompts, responses, tools, commands, edits, approvals, MCP activity and tokens in one trace.
Local-first: data lives in JSONL and goes only where you point it. You can also sign in to beacon.sh for the cloud version, and setup offers an explicit local option.
Many install paths: Homebrew on macOS, plus Windows, Linux, MDM and CI routes, or just ask your coding agent to read beacon.sh/start.md and set it up. MIT licensed.
Beacon is an open-source, local-first memory layer for coding agents. It gathers full session history from Claude Code, Cursor, Codex, OpenCode, Cline and 20+ other harnesses, then turns the useful workflows, your corrections, debugging patterns and repo conventions into knowledge that future agents can reuse. A problem solved by one agent shouldn't have to be learned from scratch by another.
Reviewed knowledge can be retrieved by later agents through MCP or Agent Skills, and you can replay any session exactly, with prompts, responses, tools, commands, edits, approvals and token use in one trace. Data is stored as durable JSONL, you choose the destinations, and no harness has you locked in.
Features
Cross-harness history: sessions from Claude Code, Cursor, Codex, OpenCode, Cline and 20+ other agents in one place.
Knowledge that compounds: workflows, corrections, debugging patterns and repo conventions that outlive a single session.
Shared agent memory: reviewed knowledge that later agents retrieve through MCP or Agent Skills.
Exact session replay: prompts, responses, tools, commands, edits, approvals, MCP activity and tokens in one trace.
Local-first: data lives in JSONL and goes only where you point it. You can also sign in to beacon.sh for the cloud version, and setup offers an explicit local option.
Many install paths: Homebrew on macOS, plus Windows, Linux, MDM and CI routes, or just ask your coding agent to read beacon.sh/start.md and set it up. MIT licensed.
