latitude-dev/latitude-llm

Latitude is the open-source AI monitoring platform.

What it solves

Latitude provides a system for making AI agents "self-healing" by automating the cycle of observing failures, understanding why they happened, and fixing them. It eliminates the manual effort of digging through logs to find agent errors and then manually implementing fixes.

How it works

Latitude uses a four-step process to maintain agent reliability:

  1. Observe: It captures full execution traces, including multi-turn sessions and tool calls, using a single line of telemetry.
  2. Understand: It automatically groups failing traces into "signals" to identify patterns of failure and their frequency.
  3. Fix: It dispatches coding agents (like Claude Code or Cursor) with the full context of the failure, including sample traces and deep links, allowing the coding agent to write a fix and open a PR.
  4. Verify: It replays the failing traces against the new fix to ensure the issue is resolved and does not regress.

Who it’s for

Developers building AI agents that require high reliability and scale, particularly those using LLM providers like OpenAI, Anthropic, and Bedrock, or frameworks like LangChain and the Vercel AI SDK.

Highlights

  • Provider Agnostic: Works with most major LLM providers and is compatible with OpenTelemetry (OTel).
  • Agent Dispatch: Integrates with coding agents to automate the actual code changes needed to fix bugs.
  • Self-Hostable: Can be deployed via Docker Compose, Kubernetes (Helm), or one-click via Railway.
  • MCP Server: All UI functionality is accessible via a Model Context Protocol (MCP) server and CLI for agentic workflows.

Related

  • Project
    traceroot-ai/traceroot

    An open-source observability and self-improving layer for AI agents that detects production failures, root-causes them to source code, and opens verified fix PRs.

  • Project
    lmnr-ai/lmnr

    An open-source observability platform for AI agents that provides tracing, real-time monitoring, and evaluation tools to debug and optimize LLM applications.

  • Project
    greyhaven-ai/autocontext

    A recursive self-improving harness that helps AI agents iteratively solve tasks by running evaluations, retaining successful lessons, and distilling stable behavior.

  • Project
    reticlehq/reticle

    A verification layer for AI coding agents that proofreads code by monitoring a running app's internal state and network calls to catch silent failures.

  • Project
    canwhite/AgentEval

    A transparent HTTP proxy that captures Agent-LLM traffic to provide automated grading, behavioral diagnosis, and source-code probing for AI agent evaluation.