Launching today
agent-lens
Run the scientific method on your LLM agent
2 followers
Run the scientific method on your LLM agent
2 followers
Most LLM agent debugging is vibes-based: edit a prompt, eyeball the output, repeat. agent-lens makes it structuralPause a running agent at any LLM call. State a hypothesis. Fork with edited messages. Get a structural diff — latency, tokens, cost — with verdict: improved/regressed.No other tool has pause, fork, or run diff. Local-first, SQLite, zero infra. pip install agentlens-tracer
