Voker

Voker, made by Invoke Labs, Inc. (Y Combinator-backed), is an analytics platform for AI agents — automatically detecting user intent, identifying when an agent fails to satisfy a request, and measuring successful resolutions, used by product teams at Lightfield, True Classic, and others. Who it's for: teams operating AI agents in production who want to know when and why the agent is failing users, before those users complain, rather than relying on manual review or support-ticket volume as the only signal. Strengths: automatically detecting agent failures ("corrections" — where a user has to redirect or correct the agent) is a genuinely specific, actionable signal beyond generic conversation logs. A real, usable free tier (2,000 events/month, 30-day retention) lets a team evaluate on real production data before paying. Weaknesses and limitations: cross-framework support (OpenAI, Anthropic, Gemini, LangChain, CrewAI, Vercel AI SDK) is broad, but real value depends on actually integrating the tracking into your agent's event stream, which takes real setup work. Real-world scenario: a team running a customer-facing AI agent wants to know automatically when it's failing to satisfy user requests, rather than waiting for users to complain or manually reviewing transcripts.