Latitude

Cesar Miguelanez and Gerard Clos started Latitude to close the feedback loop between observing broken AI agents and fixing the code that caused the failures. Instead of leaving teams to manually sift through thousands of execution logs, the platform links runtime trace telemetry directly to automated agent dispatch and regression suites. Today, it gives engineering squads an open-source control plane that tracks agent trajectories and verifies pull requests against real production payloads.

Why Latitude instead of LangSmith

LangSmith remains the default SaaS choice for teams invested in the LangChain ecosystem, providing managed tracing, prompt management, and human evaluation queues without infrastructure overhead. However, LangSmith keeps prompt data and runtime traces in a proprietary cloud, charging per seat and per trace while reserving on-premise deployments for custom enterprise contracts. Latitude shifts that equation to internal infrastructure: it runs entirely inside your VPC, giving platform teams complete data residency and zero external egress costs for sensitive customer prompts.

From an operations standpoint, adopting Latitude exchanges subscription fees for infrastructure responsibility. A production deployment requires PostgreSQL with pgvector for relational state, ClickHouse for trace analytics, Redis and BullMQ for queues, Temporal for evaluation workflows, and S3-compatible storage for span payloads. LangSmith abstracts that away behind an API key. But for DevOps engineers managing strict compliance boundaries, Latitude provides standard OpenTelemetry ingestion, direct Model Context Protocol integration, and an automated pipeline that dispatches coding agents like Claude Code or Cursor to submit pull requests for recurring trace failures.

What it does well

• Ingests traces via standard OpenTelemetry protocols and native TypeScript and Python SDKs, integrating into existing monitoring pipelines without vendor lock-in.
• Clusters runtime failures into tracked signals by error type and behavior pattern, converting unstructured log noise into actionable issue queues.
• Exposes an integrated Model Context Protocol server and CLI so coding assistants can inspect trace context directly from local development environments.
• Dispatches automated bug fixes to coding agents with full trace execution paths, prompting them to open targeted pull requests with minimal context loss.