Beacon.sh launches cross-harness agent memory.
Beacon is an open-source, MIT-licensed memory layer that captures sessions across Claude Code, Cursor, Codex, OpenCode, and 20+ other harnesses, then turns useful workflows and corrections into reusable skills. It also normalizes agent activity into inspectable telemetry for local dashboards and security infrastructure. [Beacon homepage](https://beacon.sh/) [GitHub repository](https://github.com/Asymptote-Labs/agent-beacon)
Beacon’s strongest idea is portability: agent memory should follow developers across tools instead of becoming another vendor silo. Its real test will be whether selective, reviewed memories outperform noisy transcript archives.
- –Captures prompts, tool calls, commands, file changes, approvals, MCP activity, and token usage in a shared event model.
- –MCP and Agent Skills let lessons learned in one harness transfer to others.
- –Local-first JSONL storage and MIT licensing make the telemetry layer inspectable and easier to integrate with existing security systems.
- –The same breadth creates governance risk: captured code, credentials, and agent actions demand strong retention, redaction, and approval controls.
- –Competes in an increasingly crowded agent-memory category, but differentiates through cross-harness telemetry and reusable skills.
DISCOVERED
1h ago
2026-11-02
PUBLISHED
6h ago
2026-11-02
RELEVANCE
AUTHOR
Justin D'Souza
