- vendor
- Armature
- whatItIs
- Analytics and automated-evaluation platform for AI agents and MCP servers that captures user sessions happening inside third-party AI clients (Claude Connector, ChatGPT App) rather than a developer's own UI, then scans sessions for failures, loops and dead ends.
- hosting
- managed-only
- pricingModel
- freemium usage-based: free tier with 1,000 credits/month (covering sessions or eval runs), then $50 per 1,000 additional credits, custom enterprise plans
- modelProviders
- agnostic
- stateModel
- session-based (captured from Claude Connector, ChatGPT App, or MCP client sessions)
- toolModel
- SDK + MCP integration
- includesBrowser
- false
- includesMemory
- false
- longRunning
- session-scoped
- maturity
- preview
- launched
- 2026-08-03
- notes
- Y Combinator-backed. Launch date (2026-08-03) is sourced from the founder's own Show HN post (https://news.ycombinator.com/item?id=49175461 area / HN Algolia record, title 'Show HN: Product analytics (and evals) for agent sessions on your MCP'), not stated on armature.tech itself. Distinct from tracked observability/eval tools (LangSmith, Langfuse, AgentOps, Confident AI, Galileo): those instrument code you control and trace calls you make; Armature's premise is that sessions on an MCP Connector or ChatGPT App run inside a client the developer does not own, so it captures and evaluates those out-of-band sessions instead of in-process traces. Maturity marked preview based on the site's 'launch offers' framing for dashboard features, not an explicit GA/beta label.
- observability
- none
- hitlSupport
- false
