Latitude

Latitude is an open-source observability and evaluation platform for AI agents running in production. It ingests traces of agent conversations — compatible with existing OpenTelemetry pipelines — and analyzes completed sessions to surface behavioral patterns such as user escalations, broken trust, and failed tool calls. Recurring problems are grouped into 'signals' with trend analysis and prioritization, so a team can see which failure modes are actually common rather than reacting to one-off complaints. What sets Latitude apart is its agent dispatch feature: when it detects a problem, it can hand the issue off directly to a coding agent like Claude Code or Cursor, complete with full context and sample traces, and open a GitHub pull request with a fix. Teams can also define custom signals for specific quality checks, auto-generate golden evaluation datasets from validated production traces, and search sessions with semantic and metadata filters. Alerts go out over Slack, email, or webhooks, and an MCP server lets agents interact with the workspace directly. The company, Latitude Data S.L., targets development teams building agents at scale who want production monitoring without being locked into a single vendor's proprietary format. Specific pricing isn't detailed on the site.