Description
Latitude is an open-source AI agent observability, monitoring, and LLM evaluation platform. It provides full observability into what's failing in production, helping teams discover underlying issues, get alerts when something breaks, and verify that a fix worked.
Latitude analyzes completed sessions to extract what conversations were actually about and where things went wrong, grouping recurring failures into prioritized signals with trends and failing traces attached. When a signal appears or escalates, it can dispatch a coding agent such as Claude Code or Cursor with full context, sample traces, and a deep link, so the agent implements the smallest correct fix and opens a pull request. Alerts can be sent to Slack, email, or webhooks.
The platform is OTEL-compatible with no proprietary format or lock-in, and includes custom signals, automatic golden dataset management, human annotations, advanced semantic and metadata filtering, and failure-mode clustering. It offers an MCP server, a free tier, and enterprise security including SOC 2 Type II certification.
Latitude's Core Features
Open-source AI agent observability and monitoring
Conversation intelligence from completed sessions
Automatic failure grouping into prioritized signals
Agent dispatch to Claude Code or Cursor that opens a PR
Alerts via Slack, email, or webhooks
OTEL-compatible tracing with no lock-in
Golden dataset management and human annotations
Failure-mode clustering and advanced filters
How to use Latitude?
Set up telemetry: Drop in the SDK or point your OpenTelemetry pipeline at Latitude.
Observe traffic: Let Latitude trace your agent and capture messages, cost, tool calls, and errors.
Review signals: See recurring failures grouped and prioritized with failing traces.
Dispatch a fix: Let Latitude send a coding agent with full context to open a fix PR.
Latitude's Use Cases
- Agent monitoring
- Failure triage
- Self-healing agents
- LLM evaluation
