Loading tool details…
Stay free if you only need basic features. Upgrade if you need advanced features. Most solo builders can start free.
Both do LLM tracing and prompt management. Langfuse is MIT-licensed with a self-hosted option — meaningful for regulated industries. LangSmith is deeper on LangChain-specific ergonomics and integrates tightly with LangGraph. Many teams evaluate both before committing.
No. Langfuse works with raw OpenAI/Anthropic SDKs, LlamaIndex, Vercel AI SDK, LiteLLM, Haystack, and generic OpenTelemetry. LangChain is one of many supported integrations.
Yes — the docker-compose and Helm chart run the full stack (app + ClickHouse + Postgres + Redis). Many enterprises use the self-hosted version to keep prompts and traces inside their own VPC.
You create versioned prompts in the Langfuse UI and label them (`production`, `staging`). Your app calls `langfuse.get_prompt('name', label='production')`, which returns the current version — swapping a prompt requires no redeploy.
An observation is a single logged unit — typically an LLM generation, a tool call, or a retrieval span. A single user request that fires 5 LLM calls plus 3 tool calls generates 8 observations.
Start with the free plan — upgrade when you need more.
Get Started Free →Still not sure? Read our full verdict →
Last verified March 2026