Layer 4 · self-hosting reality check
What it actually takes to self-host Langfuse
The docs say 2 GB. In practice you want 4 GB. Here is the honest version — real requirements, real monthly cost, what you will be maintaining, and the one thing that catches people out.
Usually reached from LangSmith alternatives, where Langfuse is one of the picks.
Wondering whether you need to at all? Is LangSmith free? — what the free tier actually allows, and where the wall is.
| RAM — documented minimum | 2 GB |
|---|---|
| RAM — what it really needs | 4 GB |
| CPU | 2 vCPU |
| Disk | Trace volume × retention — LLM traces are large because they contain full prompts |
| Monthly cost | $24–40/mo plus Postgres and ClickHouse, against per-trace observability pricing |
| Setup time | 2 hours |
| How you install it | docker compose — web, worker, PostgreSQL, ClickHouse, Redis, and S3-compatible storage |
| Ongoing maintenance | Moderate. Five components, and trace volume grows faster than teams expect. |
| Where it stops scaling | High trace volume once ClickHouse is doing the storage. Postgres-only mode is for evaluation, not production. |
The thing that catches people out
Traces contain your full prompts and completions, which means your observability database now holds every piece of customer data that ever passed through a model. That is a data-retention and compliance question, not just a disk one. Set retention deliberately and decide what to redact before you instrument production, not after a subject-access request.
When not to self-host Langfuse
You need one dashboard covering both application and LLM observability. Langfuse is LLM-specific by design.
Every guide here carries this section. A site that only ever tells you to self-host is selling something — the useful answer is sometimes no.
Other Layer 4 self-hosting guides
- Self-hosting Ollama8 GB VRAM for a 7B model at usable speed; 24 GB for 30B-class
- Self-hosting vLLM24 GB VRAM minimum for useful production serving
- Self-hosting QdrantVectors × dimensions × 4 bytes, in RAM, plus overhead — 1M × 768d is roughly 3 GB
- Self-hosting pgvector8 GB — the HNSW index wants to be resident
- Self-hosting LlamaIndex4 GB for the app; your vector store is the real cost
- Self-hosting faster-whisper5 GB VRAM for large-v3 in float16; 2 GB with int8
Common questions
- How much RAM does Langfuse actually need?
- 4 GB in practice. The documented minimum is 2 GB, which is the figure at which the process starts rather than the figure at which it works under real use. 2 vCPU alongside it.
- What does self-hosting Langfuse cost per month?
- $24–40/mo plus Postgres and ClickHouse, against per-trace observability pricing This is commodity VPS pricing and excludes your time, which is the larger cost for most people — budget for moderate. Five components, and trace volume grows faster than teams expect.
- How long does it take to set up Langfuse?
- 2 hours, via docker compose — web, worker, PostgreSQL, ClickHouse, Redis, and S3-compatible storage.
- When should I NOT self-host Langfuse?
- You need one dashboard covering both application and LLM observability. Langfuse is LLM-specific by design.
- What is the most common mistake when self-hosting Langfuse?
- Traces contain your full prompts and completions, which means your observability database now holds every piece of customer data that ever passed through a model. That is a data-retention and compliance question, not just a disk one. Set retention deliberately and decide what to redact before you instrument production, not after a subject-access request.