Use any AI with your data
Because your notebook is plain files on your own disk, the AI you already use can read it. ResearchOS ships a built-in AI Helper, a prompt that teaches any model how your data is structured, so it can draft entries, cross-reference protocols, and answer questions about your own work. This page covers both ways to use it, pasting into a chat assistant and pointing an agent at your folder.
The AI Helper prompt
Open Settings and find the AI Helper section. It gives you a schema-aware prompt, a block of text that explains how a ResearchOS folder is laid out (where notes, experiments, methods, and projects live, and what each file looks like). Hand that to any model and it understands your data without you explaining the structure every time.
You choose a size to fit the model you are using.
- Minimal (around 6k tokens) is for tiny windows or small local models.
- Lean (around 21k tokens) fits most chat windows.
- Full (around 42k tokens) is best for drafting on models that accept a large context window.
Flow 1: paste into a chat assistant
The simplest way. No setup, works with any chat AI, and nothing leaves your machine except the text you choose to paste.
- Copy the prompt. In Settings, pick a size and click copy, or use the one-click button to open it directly in Claude, ChatGPT, Gemini, or Copilot. (Microsoft Copilot is free with many university accounts.)
- Paste in your data or your question. Paste the prompt, then paste the note or experiment you want help with, or just ask. Because the prompt taught the model your schema, it knows what a running-log note or a PCR method looks like and can draft in the same shape.
- Bring the result back. Copy the model's draft back into the matching field in ResearchOS. You stay in control of what gets saved.
Flow 2: point an agent at your folder
The more powerful way. An agentic tool with read access to your data folder can work across your whole notebook at once, no copy-paste. Give it the AI Helper prompt plus access to the folder and it can draft entries, fill in notes, and cross-reference protocols alongside you.
This works with any tool that can read a local folder, for example:
- A coding agent like Claude Code or Cursor, opened on your data folder.
- A filesystem MCP server, so an assistant can list and read your files.
- A local model (via Ollama, LM Studio, and the like) pointed at the folder, with inference running locally. Check that any connected tools also run locally before relying on this for private data.
What it can and cannot do
An AI can read anything in the folder you give it access to, and draft content for you to place. Whether it can write files depends on the tool. A paste-flow chat AI only suggests text you copy back, while a folder-access agent can write files directly if you let it. Treat agent drafts like a collaborator's, review before you rely on them, and lean on version history so you can always see and undo what changed.