OpenLIT

About OpenLIT

OpenLIT is an Apache 2.0 open-source observability and engineering platform for LLM and agent workloads, built on OpenTelemetry rather than on a proprietary tracing format. That choice is the substance of the product: because instrumentation emits standard OTel spans, traces can go to your existing collector and backend instead of being locked into a vendor's storage, and the same pipeline that carries your service traces carries your agent traces. Coverage runs wider than the usual LLM-call span — it instruments GPUs, LLMs, MCP servers, vector databases and coding agents, which means an agent's slow step can be attributed to the retrieval layer or the GPU rather than assumed to be the model. Around tracing it adds the adjacent pieces teams otherwise assemble separately: running LLM evaluations, managing prompts, comparing models against one another, and storing API keys in a built-in vault rather than in environment variables scattered across services. The lifecycle framing on the site runs instrument, develop, manage, observe, improve, covering both development and production stages. Deployment is self-hosted via Helm chart or Docker with OAuth sign-in, documented for both. A fully hosted OpenLIT Cloud is announced for teams that would rather not operate it, but is not yet available.

Key Features

Tags