Layer 4 · self-hosting reality check
What it actually takes to self-host RAGFlow
The docs say 4 GB. In practice you want 16 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 Vectara alternatives, where RAGFlow is one of the picks.
Wondering whether you need to at all? Is Vectara free? — what the free tier actually allows, and where the wall is.
| RAM — documented minimum | 4 GB |
|---|---|
| RAM — what it really needs | 16 GB |
| CPU | 8 vCPU; a GPU materially speeds document parsing |
| Disk | 40 GB plus your corpus, and parsed artifacts are larger than the sources |
| Monthly cost | $70–120/mo — the heaviest thing on this list, and the parsing quality is why |
| Setup time | Half a day |
| How you install it | docker compose — it bundles Elasticsearch, MySQL, MinIO and Redis alongside the app |
| Ongoing maintenance | Moderate to high. Five stateful services, and Elasticsearch has opinions about memory. |
| Where it stops scaling | Large corpora with difficult documents — that is its whole purpose. Parsing throughput binds first. |
The thing that catches people out
Document parsing is the reason to choose RAGFlow and it is genuinely compute-hungry — a large PDF corpus can take days on CPU, and the process looks hung when it is working. Provision a GPU if the corpus is real, and never start a large ingest on a machine you need for something else that week.
When not to self-host RAGFlow
Your documents are clean text or markdown. RAGFlow's deep parsing is wasted on them and you are paying for it in RAM.
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 RAGFlow actually need?
- 16 GB in practice. The documented minimum is 4 GB, which is the figure at which the process starts rather than the figure at which it works under real use. 8 vCPU; a GPU materially speeds document parsing alongside it.
- What does self-hosting RAGFlow cost per month?
- $70–120/mo — the heaviest thing on this list, and the parsing quality is why This is commodity VPS pricing and excludes your time, which is the larger cost for most people — budget for moderate to high. Five stateful services, and Elasticsearch has opinions about memory.
- How long does it take to set up RAGFlow?
- Half a day, via docker compose — it bundles Elasticsearch, MySQL, MinIO and Redis alongside the app.
- When should I NOT self-host RAGFlow?
- Your documents are clean text or markdown. RAGFlow's deep parsing is wasted on them and you are paying for it in RAM.
- What is the most common mistake when self-hosting RAGFlow?
- Document parsing is the reason to choose RAGFlow and it is genuinely compute-hungry — a large PDF corpus can take days on CPU, and the process looks hung when it is working. Provision a GPU if the corpus is real, and never start a large ingest on a machine you need for something else that week.