Find where your AI agent breaks. Load testing, cost-per-resolved-task, and chaos engineering for AI agents.
Status: pre-alpha, under active development — built in public. The 0.0.1 PyPI release is a name reservation stub; nothing works yet.
- The Knee Point — the concurrency at which your agent's latency/quality cliff begins
- $/Resolved Task — spend divided by successfully completed tasks
- Resilience Score — resolution rate under injected chaos vs. clean baseline
Follow along: kneepoint.dev
MIT licensed.