[!IMPORTANT] AgentLens 1.0 is currently in Alpha. 1.0 is a Clean Rebuild centered on Canonical Observation + Evidence; 0.x serves only as a validated reference for behavior and design and does not enter the 1.0 runtime.

Why AgentLens

Codex, Claude Code, Pi, Hermes, OpenCode and other tools scatter sessions, tool calls, and capability configuration across JSONL, SQLite, Hooks, and configuration directories. AgentLens organizes these observable facts into a unified, traceable view, helping you answer:

• What did the agent just do, and at which step did it fail?
• What turns, tool calls, and lifecycle events did a long task go through?
• Does a given conclusion come from historical records, native tail-reading, or a Runtime Hook?
• Which Skills, MCPs, and Plugins are installed on this machine, and which have actually been invoked?
• Which data is complete, which has only partial coverage, and which sources are currently unavailable?

AgentLens shows behavior that its sources can prove. It does not claim to read hidden reasoning, and it does not package speculation as fact.

Get started in 30 seconds

Desktop app

Go to [GitHub…