Agent Beacon: open-source local telemetry layer records what AI agents do across 23+ harnesses
Scobleizer · x · 2026-09-11
- Agent Beacon by Asymptote-Labs is an open-source runtime telemetry/security layer for AI agents (410+ GitHub stars).
- Agents can call tools, run shell commands, modify files, and touch production systems — but when something breaks, teams often reconstruct what happened from scattered logs. Beacon records activity locally in real time: tool calls, shell commands, file changes, approval decisions, session/harness context.
- Its key design normalizes events into a single format across 23+ agent harnesses (Claude Code, Codex, etc.), so security teams reason about the underlying action itself instead of writing per-harness detection logic.
- Fully local and open source, supports local, CI, browser, and cloud; the repo ships a threat-rules spec, a sandbox self-verify skill, and a browser extension.
More from coding & agent
- Claude Code's five maintenance commands: skills, cost, context debt — addyosmani · 2026-09-11
- OpenTrustBench: A Fully Local, Zero-Telemetry MCP Server Security Scanner — BrilliantSecret143 · 2026-09-11
- Fighting AI Code Slop With Package Cohesion Before Review — newlido · 2026-09-11
- Codex drives DaVinci Resolve via MCP to auto-edit video end to end — yungcontent · 2026-09-11
- Xcode 27 Ships Built-in MCP Server: One Command Hands Debugging to Claude or Codex — rudrank · 2026-09-11
- User stumbles on mystery browser-use RL training environment, sparking agent training leak jokes — xeophon · 2026-09-11