About Latitude
Latitude is an open source observability platform for AI agents. It traces an agent running in production, groups the failures it finds into issues, and can hand those issues to a coding agent to fix.
Traces are ingested through the vendor's SDK or any OpenTelemetry pipeline, so there is no proprietary format. Latitude analyzes completed sessions to surface behavior patterns such as escalations, abandonments, retries and tool failures. Failed traces become signals, similar failures are clustered into a single issue with examples, trend and affected users, and alerts can go to Slack, email or webhooks. Users can define custom signals for any behavior they want tracked, annotate traces and spans with feedback, and filter with a mix of semantic search, exact text and metadata. Golden datasets are built from validated production traces for evals and regression tests. When an issue appears or escalates, Latitude can dispatch Claude Code or Cursor with sample traces and context to propose a fix and open a pull request, and an MCP server lets this run from the editor.
It is built for teams that ship LLM agents and need to monitor quality after release. The code is published under the MIT license.
The Starter plan is free with 20,000 credits per month, 30 day data retention and unlimited seats. Pro is $99 per month with 100,000 credits, 90 day retention and priority support, and extra credits cost $20 per 10,000. Enterprise is custom priced and adds SSO, role based access, on premises or cloud deployment and an uptime SLA.