Self-hosted AI apps that run without a GPU
Short answer
25 of the 40 apps we track can run on an ordinary CPU server according to their own docs.
The lowest official requirements: LibreChat (minimum 1 vCPU · 1 GB RAM (recommended 2 GB RAM), from €6.49/month); Langflow (minimum 1 vCPU · 1.5 GB RAM (recommended 2 vCPU · 4 GB RAM), from €11.99/month); MinerU (minimum 2 GB RAM, from €11.99/month).
Apps that need a GPU for practical use: ComfyUI, vLLM, Tabby.
Not listed because their docs do not state hardware or GPU needs: n8n, Firecrawl, Open WebUI, NextChat, PrivateGPT, PDFMathTranslate, Vane (formerly Perplexica), Qdrant, Activepieces, Paperless-AI.
Last updated · generated from the same verified data as each app page
Comparison table
| App | Category | Why it qualifies | Official hardware | Est. server cost |
|---|---|---|---|---|
| LibreChat | Private ChatGPT-style chat | Docs: no GPU needed | minimum 1 vCPU · 1 GB RAM (recommended 2 GB RAM) | from €6.49/mo |
| Langflow | AI app & agent builders | Official requirements list CPU and RAM only | minimum 1 vCPU · 1.5 GB RAM (recommended 2 vCPU · 4 GB RAM) | from €11.99/mo |
| MinerU | Document processing | GPU optional (CPU mode documented) | minimum 2 GB RAM | from €11.99/mo |
| Dify | AI app & agent builders | Official requirements list CPU and RAM only | minimum 2 vCPU · 4 GB RAM | from €19.99/mo |
| OpenHands | Coding assistants | Official requirements list CPU and RAM only | recommended 2 vCPU · 4 GB RAM (no separate minimum published) | from €19.99/mo |
| LobeHub (LobeChat) | Private ChatGPT-style chat | Official requirements list CPU and RAM only | minimum 2 vCPU · 4 GB RAM · 20 GB disk (recommended 4 vCPU · 8 GB RAM · 50 GB disk) | from €19.99/mo |
| AnythingLLM | Knowledge base & RAG | Official requirements list CPU and RAM only | minimum 2 vCPU · 2 GB RAM · 5 GB disk | from €19.99/mo |
| LiteLLM | LLM gateway & observability | Official requirements list CPU and RAM only | minimum 1 vCPU · 4 GB RAM | from €19.99/mo |
| Open Notebook | Knowledge base & RAG | Official requirements list CPU and RAM only | minimum 4 GB RAM | from €19.99/mo |
| FastGPT | Knowledge base & RAG | Official requirements list CPU and RAM only | minimum 2 vCPU · 4 GB RAM (recommended 2 vCPU · 8 GB RAM) | from €19.99/mo |
| Coze Studio | AI app & agent builders | Official requirements list CPU and RAM only | minimum 2 vCPU · 4 GB RAM | from €19.99/mo |
| Milvus | Data labeling & vector databases | Official requirements list CPU and RAM only | minimum 8 GB RAM (recommended 4 vCPU · 16 GB RAM) | from €35.99/mo |
| Label Studio | Data labeling & vector databases | Official requirements list CPU and RAM only | minimum 8 GB RAM (recommended 16 GB RAM · 50 GB disk) | from €35.99/mo |
| MaxKB | Knowledge base & RAG | Official requirements list CPU and RAM only | minimum 4 vCPU · 8 GB RAM · 100 GB disk | from €35.99/mo |
| RAGFlow | Knowledge base & RAG | Official requirements list CPU and RAM only | recommended 4 vCPU · 16 GB RAM · 50 GB disk (no separate minimum published) | from €69.99/mo |
| Langfuse | LLM gateway & observability | Official requirements list CPU and RAM only | recommended 4 vCPU · 16 GB RAM · 100 GB disk (no separate minimum published) | from €69.99/mo |
| Onyx (formerly Danswer) | Knowledge base & RAG | Official requirements list CPU and RAM only | minimum 4 vCPU · 10 GB RAM · 32 GB disk (recommended 8 vCPU · 16 GB RAM · 500 GB disk) | from €69.99/mo |
| Sim | AI app & agent builders | Official requirements list CPU and RAM only | minimum 2 vCPU · 12 GB RAM · 20 GB disk (recommended 4 vCPU · 16 GB RAM · 50 GB disk) | from €69.99/mo |
| Ollama | Local model runtimes | GPU optional (CPU mode documented) | Not published (measured by us: 1.9 GB RAM while working) | from €19.99/mo (based on our test, not official) |
| llama.cpp | Local model runtimes | GPU optional (CPU mode documented) | Not published | Not estimated (no official RAM figure) |
| whisper.cpp | Speech & translation | GPU optional (CPU mode documented) | Not published | Not estimated (no official RAM figure) |
| LocalAI | Local model runtimes | GPU optional (CPU mode documented) | Not published | Not estimated (no official RAM figure) |
| Khoj | AI search & research assistants | GPU only needed for offline chat with local models | Not published | Not estimated (no official RAM figure) |
| LibreTranslate | Speech & translation | GPU optional (CPU mode documented) | Not published | Not estimated (no official RAM figure) |
| Speaches | Speech & translation | GPU optional (CPU mode documented) | Not published | Not estimated (no official RAM figure) |
An app is listed only if its own docs say the GPU is optional or not needed, or its official requirements list CPU and RAM only. Archived projects are excluded. Running a large AI model on the same CPU server is a separate question; that depends on the model. Cost = cheapest Hetzner Cloud Regular Performance (shared AMD vCPU) plan that meets the project's official RAM and vCPU figure, list price read on 2026-10-22, excl. VAT, server only (no LLM API usage, backups or your time).
Frequently asked questions
Which self-hosted AI apps run without a GPU?
25 of the 40 apps we track can run on an ordinary CPU server according to their own docs. They are: LibreChat, Langflow, MinerU, Dify, OpenHands, LobeHub (LobeChat), AnythingLLM, LiteLLM, Open Notebook, FastGPT, Coze Studio, Milvus, Label Studio, MaxKB, RAGFlow, Langfuse, Onyx (formerly Danswer), Sim, Ollama, llama.cpp, whisper.cpp, LocalAI, Khoj, LibreTranslate, Speaches.
What is the cheapest self-hosted AI app to run on a CPU server?
LibreChat: official minimum 1 vCPU · 1 GB RAM (recommended 2 GB RAM), which fits a Hetzner Cloud CPX02 at about €6.49/month (server only).
Which self-hosted AI apps need a GPU?
ComfyUI: The README claims it can run even the largest open models on as little as 4 GB VRAM plus 8 GB RAM using weight streaming. NVIDIA, AMD, Intel, Apple Silicon and Ascend are supported. vLLM: NVIDIA: compute capability 7.5 or higher (e.g. T4, RTX 20xx, A100, L4, H100, B200). Tabby: NVIDIA GPU recommended. CodeLlama-7B in int8 needs about 8 GB VRAM. int8 needs compute capability 7.0+ (or 6.1), float16 7.0+, bfloat16 8.0+.
Can I run a local LLM without a GPU?
Ollama: NVIDIA GPUs need compute capability 5.0+ and driver 550+ (570+ for compute capability 5.0–6.2). CPU-only also works. Memory needs depend on the model you load, not on Ollama itself. LocalAI: Optional. Images exist for NVIDIA CUDA 12/13, AMD ROCm, Intel oneAPI and CPU-only. No official RAM figures. Needs depend on the models you load. Speaches: Optional (CUDA image). A CPU image is available. No official RAM figures.
Related
Every figure on this page comes from the linked app pages, which cite the project's own repository and docs. Spotted something wrong? Tell us.