Repolode
latitude-dev

latitude-dev/latitude-llm

Open-source observability for AI agents that groups failing traces, dispatches coding agents to fix them, and replays fixes against real traces.

Open-source alternative to:LangSmithLangfuseHelicone

Stars
4,730
Forks
401
Open issues
109
Last commit
6 hours ago

What is it?

Latitude is an open-source observability platform for LLM applications and AI agents. Telemetry is added with a short SDK setup or OpenTelemetry, and every LLM call, multi-turn session and tool call is recorded as a trace.

Failing traces are grouped automatically into tracked signals that show status, volume and trend. From a signal, Latitude can dispatch a coding agent such as Claude Code or Cursor with the relevant traces and context, and that agent can write a fix and open a pull request. Fixes are then replayed against the original failing traces through regression datasets and monitors.

The project is written in TypeScript and provides a UI, an MCP server and a CLI. It can be self-hosted with Docker Compose or Kubernetes via a Helm chart, and Python is supported alongside TypeScript.

Key features

  • 01Captures LLM calls, multi-turn sessions and tool calls as traces via SDK or OpenTelemetry.
  • 02Groups failing traces automatically into signals with status, size and trend.
  • 03Dispatches coding agents like Claude Code and Cursor with traces and context to fix issues.
  • 04Replays fixes against real failing traces with regression datasets and monitors.
  • 05Provides an MCP server and CLI so the same actions are available from a coding agent.
  • 06Can be self-hosted via Docker Compose or Kubernetes Helm chart.

Use cases

  • Finding out which user requests or agent workflows are failing in a production LLM app.
  • Letting a coding agent pick up recurring agent failures and open a fix PR.
  • Preventing previously fixed agent failures from returning through regression tests.
  • Running a self-hosted observability stack for AI agents on your own infrastructure.

Quick start

  1. 1

    Sign up at latitude.so and get your API key and project slug

  2. 2

    Install the TypeScript telemetry package

    npm install @latitude-data/telemetry
  3. 3

    Initialize the Latitude client with your LLM SDK as shown in the README example

Things to know

  • Full use needs a Latitude account and API key, and the free tier has a 30-day data retention limit.
  • Self-hosting requires running the Docker Compose or Kubernetes deployment yourself.

Topics

FAQ

What is latitude-llm?
Open-source observability for AI agents that groups failing traces, dispatches coding agents to fix them, and replays fixes against real traces. Latitude is an open-source observability platform for LLM applications and AI agents. Telemetry is added with a short SDK setup or OpenTelemetry, and every LLM call, multi-turn session and tool call is recorded as a trace.
Is latitude-llm free to use? What license does it have?
latitude-llm is open source under the MIT license. Check the license terms before using it in commercial products.
Is latitude-llm still maintained?
Actively maintained. The last commit to latitude-llm was on Oct 21, 2026.
What can I use latitude-llm for?
Finding out which user requests or agent workflows are failing in a production LLM app. Letting a coding agent pick up recurring agent failures and open a fix PR. Preventing previously fixed agent failures from returning through regression tests. Running a self-hosted observability stack for AI agents on your own infrastructure.
How do I install latitude-llm?
Sign up at latitude.so and get your API key and project slug Install the TypeScript telemetry package: npm install @latitude-data/telemetry Initialize the Latitude client with your LLM SDK as shown in the README example
Where can I get latitude-llm?
The source code and installation instructions are on GitHub: https://github.com/latitude-dev/latitude-llm

Summary generated by AI from the project's README. GitHub stats refresh daily.

Related projects