Deploy Langfuse on Your VPS
Deploy Langfuse on your VPS in one click. Open-source LLM observability — trace every prompt, completion, latency, and token cost, on infrastructure you own.
Langfuse is where you find out what your AI app actually did. Every call it makes is recorded as a trace: the prompt that went out, the completion that came back, how long it took, what it cost in tokens, and where it errored. When a user reports a bad answer, you open that exact trace and read the chain that produced it instead of guessing from logs. On top of tracing it does prompt management, evals, and datasets, and it works with the OpenAI SDK, LangChain, LlamaIndex, or anything that speaks OpenTelemetry.
This is the full v3 stack, wired together for you: PostgreSQL for projects and prompts, ClickHouse for the trace analytics, Redis for the ingestion queue, MinIO for raw event storage, the web app, and the background worker that folds events into ClickHouse. Your admin account is created during the deploy, so you sign in with the email and password shown in the wizard — and because that account already exists, public sign-up ships disabled. Nobody else can register on an instance that holds your prompts and completions.
What's included
- Web app + worker + PostgreSQL + ClickHouse + Redis + MinIO, wired together
- Admin account and a starter project with API keys created during deploy
- Public sign-up disabled by default — invite teammates from Settings
- Works with the OpenAI SDK, LangChain, LlamaIndex, and OpenTelemetry
- Traces, prompt management, evals, and datasets on storage you own
Common use cases
- Debugging a bad LLM answer by replaying the exact trace behind it
- Watching token cost and latency per feature before the bill arrives
- Keeping prompts and completions on your own server for privacy or compliance
Frequently asked questions
How much RAM does it need?
Around 7GB. It is six services, and ClickHouse is the hungry one — it is a real analytics database, which is what makes trace queries fast at volume. Pick a VPS with at least 8GB.
Can other people sign up on my instance?
No. Your admin account is provisioned during the deploy, so sign-up ships disabled — otherwise anyone who found the URL could register and read your prompts and completions. Invite teammates from Settings once you are in.
Does multimodal media upload work?
Not in this template. Attaching images or audio to traces needs an object-storage endpoint your browser can reach, and this stack deliberately keeps storage internal. Tracing, prompt management, evals, and datasets are all unaffected.
Why version 3 and not 2?
The v2 line is dormant — it stopped getting releases in late 2025. v3 is the only line still shipping, and it is what this template deploys.
Ready to ship?
Deploy Langfuse in under a minute.
Connect your VPS, pick this template, fill in any required fields. Vessl handles the container, SSL, and persistent storage.
Start for FreeNo credit card · BYOS · IDR billing