Catch what will break next in your AI agent.

Latitude is the open-source platform for monitoring AI agents in production. Capture agent traffic, surface failures, and turn them into trackable issues before they escalate.

Observability

Every interaction becomes a trace of spans — LLM calls, tool calls, retrieval steps — with tokens, cost, latency, and errors. Group traces into sessions for full multi-turn reviews.

Find behaviors across your traces. Semantic search (“hand off to human”), exact text, or metadata filters (model, release, user, tag). Save searches to monitor cohorts over time.

Issues

Recurring failure patterns clustered from your traces, with examples, trends, and lifecycle states. Surfaced by flaggers, evaluations, or your own annotations — then triaged, resolved, or turned into ongoing monitors.

Universally compatible

Works with any model provider, framework, or agent harness. Drop in our TypeScript or Python SDK, or send OpenTelemetry directly — OpenAI, Anthropic, Vercel AI SDK, LangChain, and Claude Code all supported.

New · Model Context Protocol

AI-native.
Your code agent, on call.

Connect your code agent to Latitude over MCP. It investigates traces, triages issues, and sets up monitors — without leaving your editor.

Claude CodeMCP

>Why is LAT-1284 spiking?

I'll take a look.

⏺latitude.investigate("LAT-1284")

1,204 traces · retry loop on get_user_orders

⏺latitude.create_monitor("retry > 3")

✓ Alerts in #ai-on-call

Root cause: 429 from upstream, no backoff. Monitor is live.

PlansOr run it yourself — open source under MIT

Starter

Free

Cloud, for trying it out

20K credits / month
30 day retention
Community support