What it does
OpenLIT is an AI observability platform and SDK for tracing model, agent, and vector-database activity. The SDK instruments supported libraries and sends telemetry over OpenTelemetry to an OpenLIT dashboard and receiver. Teams can self-host the dashboard and its ClickHouse storage, or use separate Enterprise features; the public pricing page says a hosted OpenLIT Cloud option is not yet available.
The OpenLIT quick start installs the SDK for Python or TypeScript, initializes it with an OTLP endpoint, and instruments an AI application. The guide shows supported integrations and describes cost, token, and model-call telemetry. Its default spans capture prompt and completion content; pass the documented capture_message_content=False setting when that content should not be sent. For a first check, connect one development application to a local or self-hosted OpenLIT receiver and inspect a model call before expanding instrumentation. The self-host guide describes a dashboard and OTLP receiver paired with ClickHouse. The tool detects and records behavior; the application team still has to interpret traces, tune what it captures, and fix model or application issues. The repository is Apache-2.0-licensed and the pricing page describes self-hosted OpenLIT as free. The same page distinguishes Enterprise features and says hosted Cloud is coming soon, so do not treat the self-host license as proof that every hosted or enterprise capability is free.
Strengths
- The SDK can automatically instrument supported LLMs, vector databases, MCP, and framework integrations.
- The quick start connects telemetry to an OTLP endpoint and exposes configuration for excluding prompt/completion content.
- A self-hosted deployment includes an OpenLIT dashboard and receiver with ClickHouse storage.
Watch out for
- Prompt and completion content is captured by default unless the documented option disables it, so teams should review privacy before enabling it.
- Self-hosting requires operating the dashboard, receiver, and ClickHouse storage.
- The pricing page says the hosted Cloud option is coming soon; it does not provide a current hosted plan to deploy.
Best for
- Teams instrumenting LLM, agent, or vector-search applications with OpenTelemetry.
- Developers who want model-call traces and token or cost context in an observability workflow.
- Operators who can self-host the dashboard and deliberately control prompt data collection.
Sources
- OpenLIT AI Observability Platform (Oct 25, 2026)
- Get started with AI Observability (Oct 25, 2026)
- Self-Host OpenLIT (Oct 25, 2026)
- openlit/openlit (Oct 25, 2026)
- Apache License 2.0 (Oct 25, 2026)
- OpenLIT Pricing (Oct 25, 2026)