Kelet

Kelet is an observability and debugging service for LLM applications and AI agents in production. It reads an agent's traces and signals, groups failure patterns across many sessions, identifies the root cause with supporting evidence, and generates a prompt patch to fix it. It then runs the patch against real sessions to show reliability before and after. It is built for engineering teams that already ship agents and spend time scrolling traces to guess at fixes. The dashboard shows open issues and agent health, an issues view lists each classified failure pattern with its evidence, and a sessions view lets users inspect any trace to see which step broke. It connects through OpenTelemetry, Langfuse, or its SDK, and lists compatibility with OpenAI, Anthropic, LangChain, Google ADK, pydantic, the AI SDK, CrewAI, Strands, Agno, Mastra, Mixpanel, and PostHog. Pricing has three tiers. Starter is free with 500 sessions a month and 15 days of data retention. Startup is listed at $400 a month with 5,000 sessions included, usage based pricing above that, and 30 days of retention, though it is free during early access. Enterprise is custom and adds unlimited sessions, SSO and SAML, an SLA, and dedicated support. Root cause analysis and prompt patch generation are included on every tier, and no credit card is required to start.