Latitude

Turn AI agent behavior into issues, tests.

What is Latitude? Latitude is an open-source agent observability platform for teams running AI and LLM-based agents in production. It captures rich, agent-native traces across sessions, tool calls, and reasoning steps, then turns live traffic and human judgment into scores, issues, and evals so teams can see what is breaking and why. It is aimed at engineering, product, and reliability teams that need a closed-loop system to monitor, evaluate, and continuously improve agents after they ship. Key Features: Agent-Native Observability: Records full multi-step traces with tool calls, reasoning turns, inputs, outputs, errors, and token usage so engineers can pinpoint exactly where an agent run failed. Conversation Intelligence: Analyzes completed sessions to classify what the conversation was about and flag events like escalations, resolutions, abandonments, retries, trust breaks, and tool failures. Session Search and Advanced Filters: Offers semantic search across 100% of traces, combined with exact text and metadata filters to quickly isolate specific cohorts and behaviors. Automatic Issue Discovery and Failure Clustering: Groups similar failing traces into named issues with examples, trends, affected users, and lifecycle states, plus alerts over Slack, email, or webhooks.