A model changed under your agent. Here's what broke.
You will not hear from us unless something changed. That is the whole product. agentcheck watches the models your agent runs on; when a provider ships a new model or alias it re-runs your checks and emails one forwardable diff. Every night it replays your production traces and groups regressions by cause — prompt, model, tool or flake — and runs an adversarial pack (injection, exfil, EU AI Act Art. 5). No dashboard to watch, no graphs to interpret. Uptime checks, a public badge and status page are free for one target, forever.
Tell us which model your agent runs on. When the provider ships, we re-run your suite — with the new model where your endpoint takes an override header — and email what changed. Pro.
Paste a Langfuse/LangSmith export or call agentcheck_record from your agent. Promote a trace and it replays nightly; failures are re-run before they count and grouped by cause.
http, MCP (Streamable HTTP) or A2A. GitHub login → paste a URL → three proposed checks → live badge in under two minutes. Exact/schema goldens never cost an LLM call.
Pricing
- 1 target
- every 60 min
- 7-day history
- 3 targets
- every 5 min
- 30-day history
- 10 targets
- every 5 min
- 90-day history
- 30 targets
- every 1 min
- 365-day history
Machine-readable: /pricing.json. Add-ons are per run.