Project issues include 16 GB deployments reaching memory pressure, and complex document/image/graph workloads can create spikes far above idle server use.
Large PDFs, OCR-heavy corpora and GraphRAG users.Requisiti hardware di RAGFlow: 16 GB di RAM, CPU, disco e IA locale
Scopri i requisiti ufficiali di RAGFlow: 4 core, 16 GB di RAM e 50 GB di spazio su disco, oltre alle opzioni hardware per modelli locali e ZimaOS.
RAGFlow requirements at a glance
RAGFlow publishes one of the clearest hardware floors in the ZimaOS AI catalog: x86 CPU with at least 4 cores, at least 16 GB RAM and at least 50 GB disk. Current Quickstart also requires Docker 24+, Docker Compose 2.26.1+, Python 3.13 and vm.max_map_count of at least 262144 for the Elasticsearch path.
- CPU
- Official minimum: 4 x86 CPU cores.
- RAM
- Official minimum: 16 GB. This is a real project prerequisite, not a third-party recommendation.
- Disk
- Official minimum: 50 GB. In practice, documents, parsed assets, indexes, Docker layers and model caches can require much more.
- Kernel/system setting
- vm.max_map_count must be at least 262144 for the Elasticsearch-based deployment; otherwise RAGFlow can fail to connect to ES.
- GPU
- RAGFlow server itself can use remote model APIs without a GPU. Local chat/embedding models can be attached through Ollama, Xinference, vLLM and other backends, which have separate GPU/VRAM requirements.
- Best Zima starting point
- ZimaCube 2 Pro is the first current stock Zima configuration that cleanly clears 16 GB RAM and 50 GB disk without relying on nearly-full eMMC. ZimaBoard 2 1664 is conditional on adding external SSD storage.
From official requirements to the right setup
RAGFlow sizing starts with the official 4-core/16-GB/50-GB floor before any model inference is added.
-
Official requirements
Reject hosts below the official minimum first: fewer than 4 x86 cores, under 16 GB RAM or under 50 GB usable disk are not clean RAGFlow targets.
-
Confirm your needs
Confirm host/kernel compatibility. RAGFlow requires vm.max_map_count ≥ 262144 for Elasticsearch and uses a multi-container Docker stack, so host-level settings matter.
-
Leave room to grow
Separate RAGFlow from the model provider. Remote LLM/embedding APIs keep GPU inference off-host; local Ollama/Xinference/vLLM adds another substantial RAM/VRAM workload.
-
Run it on ZimaOS
Run real document parsing and retrieval tests. Complex PDFs, OCR, GraphRAG/entity work and large corpora can push memory/CPU/disk far above the basic startup minimum.
Check every playback client
- x86 CPU with at least 4 cores
- At least 16 GB RAM
- At least 50 GB usable disk after OS/app overhead
- vm.max_map_count ≥ 262144
- Document corpus size and PDF/image complexity
- Elasticsearch versus other supported retrieval components
- Remote versus local embedding/chat models
- GraphRAG and heavy parsing features
Official minimum requirements
RAGFlow's current Quickstart publishes explicit prerequisites.
Unlike many AI frontends, these numbers should be treated as real deployment gates. A machine with exactly 16 GB RAM and 50 GB disk has little growth margin once parsing, Docker, indexes and local models are added.
| Requirement | Official minimum | What this supports |
|---|---|---|
| CPU | ≥ 4 cores (x86) | Current official Quickstart prerequisite. |
| RAM | ≥ 16 GB | Current official Quickstart prerequisite. |
| Disk | ≥ 50 GB | Current official Quickstart prerequisite. |
| Docker | ≥ 24.0.0 | Current official Quickstart prerequisite. |
| Docker Compose | ≥ v2.26.1 | Current official Quickstart prerequisite. |
| vm.max_map_count | ≥ 262144 | Required for the Elasticsearch component in the current deployment. |
When to upgrade your hardware
Upgrade RAGFlow when parsing/indexing and local-model workloads exceed the already-high minimum.
16 GB is exhausted during parsing, GraphRAG or indexing
The 50 GB floor leaves too little room for indexes and Docker data
Local LLM/embedding inference is added on the same server
50 GB is only the project minimum. Parsed images, search indexes, Docker layers, uploads and model caches can quickly make a near-minimum disk impractical.
Long-lived document libraries and local AI stacks.RAGFlow officially integrates local model servers such as Ollama/Xinference/vLLM. Their model RAM/VRAM is separate from RAGFlow's 16 GB application floor.
Fully local/offline RAG deployments.Plan hardware growth with confidence
Scale RAGFlow by adding application RAM/storage first, then model compute as a separate layer.
Give RAGFlow more than the bare 16 GB when possible
16 GB is a minimum, not a comfortable large-corpus recommendation. More memory helps absorb parser/index/database peaks and co-hosted services.
A 64 GB Creator Pack gives substantial system-memory headroom, but GPU is only useful for compatible local models.Use SSD/NVMe for Docker, indexes and parsed assets
RAGFlow depends on several persistent services and frequent index/database access. A nearly full eMMC volume is a poor match to its 50 GB floor.
Use a dedicated SSD/NVMe or ZimaCube system storage.Use remote models if the server has no suitable GPU
RAGFlow can call remote model providers and does not require a local GPU merely to run the RAG application.
This keeps the server requirement near the RAGFlow application/database tier.Add Creator Pack only for independently verified local AI
Local Ollama/Xinference/vLLM can use GPU/CUDA, but model VRAM requirements remain model-specific.
Verify RTX PRO 2000 VRAM and backend support before promising a local-model tier.Can it run on ZimaOS?
RAGFlow is currently available in the ZimaOS App Store, but its official prerequisites remain hard constraints.
Install RAGFlow only on a host that clears the official floor
The ZimaOS App Store presence does not make 8 GB hardware a valid RAGFlow recommendation.
Open RAGFlow in the ZimaOS App StoreVerify vm.max_map_count
The Quickstart requires 262144 or higher for Elasticsearch. A low value can cause 'Can't connect to ES cluster' errors that look like application failure.
Read RAGFlow QuickstartAttach local models only after sizing them separately
RAGFlow supports Ollama, Xinference, vLLM, SGLang and other local model servers, but delegates their hardware sizing to those runtimes.
Read RAGFlow local-model deploymentChoose Zima hardware for RAGFlow
RAGFlow's official 16 GB RAM and 50 GB disk floor eliminates the 8 GB Zima configurations from clean recommendations.
Do you only need RAGFlow with remote model APIs, or also local model inference?
RAGFlow with remote model/API backends
Use at least 16 GB RAM and enough SSD beyond the 50 GB floor.
- Conditional compact host with added SSDZimaBoard 2 1664
- First clean stock all-in-one RAGFlow fitZimaCube 2 Pro
RAGFlow plus same-machine local NVIDIA models
Use additional system RAM and a verified GPU whose VRAM fits the selected embedding/chat models.
- Conditional local-AI RAGFlow platformZimaCube 2 Creator Pack
Meeting the minimum does not guarantee parsing speed, corpus size, GraphRAG latency or local-model performance. Document complexity, index backend, OCR, embeddings and LLM inference all matter.
| Zima hardware | Best for | Example workload | Core configuration | Recommended boundary | Next step |
|---|---|---|---|---|---|
| ZimaBoard 2 1664 | A compact RAGFlow host only when external SSD storage is added. | RAGFlow minimum-class CPU/RAM with remote model providers. |
|
64 GB eMMC is too close to the official 50 GB disk floor once ZimaOS, Docker and data are included; add substantial SSD storage. | Get Now |
| ZimaCube 2 Pro | The first current stock Zima configuration that cleanly fits RAGFlow's 16 GB RAM and 50 GB disk minimum. | RAGFlow with remote LLM/embedding APIs, document parsing and a broader self-hosted stack. |
|
16 GB is still the project minimum, so heavy GraphRAG/OCR or large corpora may benefit from more RAM. | Get Now |
| ZimaCube 2 Creator Pack | RAGFlow with more system-memory/storage headroom and a local NVIDIA inference path. | Large document workflows plus compatible local chat/embedding models when their VRAM needs fit the actual GPU. |
|
Do not treat RTX PRO 2000 as sufficient for every local model. Verify exact GPU VRAM and backend/CUDA compatibility. | Get Now |
What the Press Says
Highlights from trusted reviewers worldwide.
“ZimaCube 2: Not just another NAS, tested with 25TB storage, local AI agents, 4K transcoding, and real homelab workflows.”Read full review
“The ZimaBoard 2 is a compact x86 server board that can be turned into a mini NAS, home server, media box, or self-hosting hub.”Read full review
“ZimaCube 2: A modern, high-performance NAS with plenty of room to grow—built for users who want more than basic storage.”Read full review
“Coverage focused on ZimaCube 2's open hardware foundation, no monthly fee, and self-hosting flexibility.”Read full review
Loved by the Community
Stories and reviews from people who build with Zima every day.
★★★★★
Zima Blade Little yet Powerful
Maybe I am not digital natives but I live with PCs since 12 years old in 1984 when IBM PC clone come to my home. Many years have passed and many operating system I've tried. For me Zima blade and CasaOS was a quantum leap for home PC enthusiast and server lab machine to make me stay curious and relevant for this era.
★★★★★
Very good!!
I use ZimaCube Pro as 5th Proxmox cluster node. It runs several VMs and containers, including a VM with GPU passthrough to run a self-hosted LLM. A specific LXC container runs a Samba server for NAS capabilities using four of six RAID 6 SATA HDDs with ZFS.
★★★★★
Great innovation for mini server!
It is very useful and makes a powerful mini server for many purposes, including university and college students in engineering and electronics. Thank you so much for making this server.
★★★★★
Avaliação ZimaBoard 2
Construí um servidor de uso pessoal. O desempenho está muito bom e funciona perfeitamente onde quer que eu esteja. A surpresa é não dependermos de grandes estruturas para termos nosso próprio servidor de dados. Como iniciante, estou gostando bastante do ZimaOS, pois ele é simples e eficiente.
Frequently asked questions
FAQ topics follow query fan-out around the 16 GB minimum, 50 GB disk, Elasticsearch memory-map errors, local GPU need, 16 GB OOM reports and RAGFlow versus model-server sizing. GitHub/Reddit are used for practical failure modes, not to replace official prerequisites.
Does RAGFlow really require 16 GB RAM?
Yes. The current RAGFlow Quickstart explicitly lists RAM ≥16 GB as a prerequisite.
Can RAGFlow run on 8 GB RAM?
It may be possible to force older/slim/custom configurations in some cases, but 8 GB is below the current official requirement and should not be a Zima hardware recommendation.
Why does RAGFlow require at least 50 GB disk?
The deployment includes multiple Docker services, search/index data and parsed document assets. Fifty GB is only the official floor; real document libraries and model caches can need much more.
Why does RAGFlow say it cannot connect to Elasticsearch?
One common host-level cause is vm.max_map_count below 262144. RAGFlow's Quickstart explicitly warns that forgetting this setting can produce 'Can't connect to ES cluster' errors.
Does RAGFlow need an NVIDIA GPU?
No for the core RAGFlow server if you use remote LLM/embedding APIs. A GPU becomes relevant when you deploy local chat/embedding models or other CUDA-accelerated components.
Why can a 16 GB RAGFlow server still run out of memory?
Sixteen GB is the minimum. Complex DOCX/PDF parsing, images, GraphRAG, Elasticsearch and concurrent tasks can create higher peaks; project issues document real 16 GB environments hitting memory problems.
Can ZimaBoard 2 1664 run RAGFlow?
CPU and RAM meet the published floor, but its 64 GB eMMC is too tight against a 50 GB minimum after ZimaOS/Docker overhead. It should only be recommended with substantial external SSD/NVMe storage.
When is ZimaCube 2 Creator Pack useful for RAGFlow?
When you also want compatible local LLM/embedding inference or need much more system RAM. The 64 GB RAM and 1 TB storage are useful, but exact GPU VRAM must still be matched to the selected model.
What sources and further reading informed this RAGFlow hardware guide?
RAGFlow's current Quickstart is the authority for 4 x86 cores, 16 GB RAM, 50 GB disk and vm.max_map_count. Its local-model guide establishes that Ollama/Xinference/vLLM inference is a separate layer. Project issues and Reddit RAG discussions were used for query fan-out around 16 GB pressure and local-model hardware. ZimaOS confirms the current AI app.
