Private document Q&A over your company's knowledge (Onyx)

Components

• Onyx (formerly Danswer): Connectors, indexing, chat and search with sources
  Official hardware: minimum 4 vCPU · 10 GB RAM · 32 GB disk (recommended 8 vCPU · 16 GB RAM · 500 GB disk) · License: MIT (Community Edition) + Onyx Enterprise License (ee directories) (Open core)
• LiteLLMoptional: Gateway to hosted model APIs with spend tracking
  Official hardware: minimum 1 vCPU · 4 GB RAM · License: MIT (core) + LiteLLM Enterprise License (enterprise directory) (Open core)
• Ollamaoptional: Local models instead of a hosted API
  Official hardware: Not published (measured by us: 1.9 GiB RAM while working) · License: MIT (Permissive)

Combined hardware and monthly cost

How we got this: we added up the official minimum vCPU and RAM each project publishes (linked on its app page) and took the cheapest Hetzner Cloud Regular Performance (shared AMD vCPU) plan that meets the total. Prices read from www.hetzner.com/cloud/regular-performance on 2026-10-22. "From" monthly price cap shown by Hetzner, excluding VAT. Prices vary by location. Covers the server only; it excludes LLM API usage, backups, extra storage, GPUs and your time. Running everything on one server is our simplifying assumption. Measured figures, where we have them, are on each app page.

Cost vs Notion AI

Deployment outline

1. Install Onyx with the official install script (Docker Compose). Choose standard mode; lite mode cannot index documents. source
2. Size the disk before you start indexing (32 GB plus about 2.5x your indexed data). source
3. In the admin panel, configure a language model provider. source
4. Optional: point Onyx at a LiteLLM proxy as its model gateway. source
5. Optional: use local models through Ollama. source
6. Add connectors for the apps and drives you want indexed. source

This is a summary of the official docs, not a full tutorial. Each app page has the exact official install command.

Gotchas

• Onyx (formerly Danswer): OpenSearch puts indices into read-only mode at 95% disk usage, which blocks all writes. Monitor disk space. source
• Onyx (formerly Danswer): Lite mode needs under 1 GB of memory but cannot index documents. source
• LiteLLM: Without a database, budgets are not enforced. max_budget is silently ignored apart from a one-time startup warning, and virtual keys do not work. source
• LiteLLM: Memory is not a usable autoscaling signal for the proxy; scale on CPU instead. source
• Ollama: Older NVIDIA cards (compute capability 5.0–6.2) need driver 570 or newer. source
• Ollama: With the Vulkan backend, Ollama needs root or the cap_perfmon capability (sudo setcap cap_perfmon+ep /usr/local/bin/ollama) to read free VRAM; otherwise it schedules models using approximate sizes. source

What this does not replace

• None of these is a docs or wiki editor. They add AI search and chat on top of your documents; they do not replace Notion itself.
• AI meeting notes and in-editor writing help are not covered.
• You connect your own model (API or self-hosted).

Other stacks

• Private ChatGPT for a small business (Open WebUI + LiteLLM)
• Self-hosted AI workflow automation (a Zapier replacement with n8n)
• Self-hosted meeting transcription (Speaches + Open WebUI)
• Self-hosted coding assistant for your dev team (Tabby)
• Self-hosted customer support AI chatbot (Dify)
• Fully offline private AI chat (Ollama + Open WebUI)

Not sure which app fits? Try the 5-question wizard.