# DocGraph — Privacy-first, AI-native personal knowledge graph > Privacy-first, AI-native personal knowledge graph. Your notes stay plain files on your machine — no cloud, no account, and the AI can run entirely on your machine. Free to use. Source: https://docgraph.app/ # Privacy-first, AI-native personal knowledge graph Your notes stay HTML and Markdown files on your machine, linked into a graph — and, when you turn it on, read by an AI that runs on this computer. It reads your PDFs, spreadsheets and Word documents too, with no import step — and nothing from your notes leaves your machine unless you choose. [Get DocGraph](https://docgraph.app/download) [Read the docs](https://docs.docgraph.app) ## Six pillars, one engine. Your knowledge graph and local RAG in one privacy-first engine, where it all finally clicks together. ### Personal knowledge graph The graph is the core, not an add-on. See how your ideas connect. ### Privacy-first Privacy isn’t a setting, it’s the architecture. You need no account and no cloud, and your notes are plain files on your disk. ### AI-native (Free & Local) AI is built into writing, search and chat. It runs free on your own Mac — with Apple’s on-device model, a built-in model or Ollama — or bring your own key. ### Local RAG Add your files and notes, ask questions, and get answers grounded in your own documents, with sources: PDFs, spreadsheets, Word files and notes, no importing required. With a local model nothing leaves your machine, so you can use AI on sensitive or confidential documents with complete privacy. ### First-class HTML, full Markdown HTML-first notes, built with real structure. Typed tables, live charts, collapsible sections, diagrams and more, all in a well-defined format that’s easy for you to read and for AI to understand. That’s why more people are moving their specs, plans and notes from Markdown to HTML. And Markdown is fully supported too. - [Why a Claude Code engineer uses HTML as specs](https://www.youtube.com/watch?v=Qrpm7E80wQ0 "Why a Claude Code engineer uses HTML as specs") - [HTML is the new Markdown](https://www.lennysnewsletter.com/p/how-i-ai-html-is-the-new-markdown "HTML is the new Markdown") - [HTML is the New Markdown for AI Agents](https://todatabeyond.substack.com/p/according-to-anthropics-engineer "HTML is the new Markdown for AI agents") ### Byte-for-byte editing DocGraph changes only the bytes you actually edited. Built in Rust, it splices just the bytes you changed and leaves every other byte untouched. Your formatting survives, and your Git diffs stay one line long. Written in Rust HTML and Markdown, side by side ## Your personal knowledge graph and local RAG in one privacy-first engine, where it all finally clicks together. Why both formats Write in Markdown when it's prose. Reach for HTML when your document needs real structure. Markdown is lovely for prose, and DocGraph edits it natively. HTML carries what a document needs once it stops being an essay: typed tables with relations and computed values, live charts, collapsible sections, diagrams and whiteboards beside the prose that explains them. Either way you edit the real file — never flattened, never converted. That structure is also what makes a note unambiguous to a machine, which is why teams building with AI agents are moving their specs and reviews to HTML too. [Why HTML — the argument and the sources](https://docs.docgraph.app/getting-started/what-is-docgraph/#why-html-first) Who it's for 01 For research & writing Ask your own notes, and get answers that cite their sources. The graph surfaces connections you forgot you made, backlinks show every note that references this one, and the local AI answers from your vault with citations — tuned to say it couldn’t find anything rather than invent an answer. 02 For business Your documents never leave the building — because they never leave the machine. Contracts, plans and internal docs stay plain files on disk: no account, no telemetry, no cloud unless you add one, and network egress is default-deny. Secret detection flags stray API keys, per-note encryption locks the ones that matter, and the app is free — nothing to procure. 03 For ideation Think on an infinite canvas, next to the notes that sparked the idea. Whiteboards live inside your vault: sketch with the pen, arrange text cards and shapes, link in your notes and images, even design a database schema with ERD tables — and every board is indexed and searchable alongside your writing. AI-native ## AI woven through the app, not bolted on the side. It runs on your machine, so every question is free. ### Ask your whole vault Answers cite the notes they came from, follow up naturally, and narrow to a tag, a person or a date range. ### Find and open “Open the note about my Berlin trip” — DocGraph finds it, and opens it when the match is clear. ### An agent that does the work Writes pages, fills tables, draws diagrams, adds todos and reorganises folders — every change asks first (or once per chat, if you allow it), and most can be undone. ### Writing help in the page /write, /ask, /summarize, /continue, Beautify, rewrite a selection, spell-check and optional ghost-text completions. ### Search by meaning Keyword and semantic search together, re-ranked by a small local model — in English or many languages. ### Works with your agents A built-in MCP server lets Claude Code, Codex, Cursor and others use your vault — off by default, and writes ask for your approval. Free AI ## Talk to your notes. Pay nothing per question. Ask your vault anything, as often as you like. The model runs on your own computer, so there’s no API key, no subscription and no token bill, and it keeps working offline. Three ways in, all free Mac ### Apple’s on-device model Already on your Mac with Apple Intelligence. One click, nothing to download. Needs macOS 26 on Apple Silicon; answers questions only. Mac ### Built-in model DocGraph downloads a small open model and runs it for you. Nothing else to install. Mac ### Ollama Already use Ollama? Pick any model you’ve installed, or let DocGraph pull its recommended one. Opt-in · not free ### Your own providers and agents Want a frontier model? Add your own API key for OpenAI, Anthropic, OpenRouter or any OpenAI-compatible endpoint, or hand a question to an agent already installed on your machine — Claude Code, Codex, Gemini CLI or OpenCode. These send your question off this computer, and you pay that provider directly, never us. Every answer shows where it ran. ## Privacy isn't a setting. It's the architecture. Your vault is just files, and it stays that way. No proprietary format, no lock-in, no account — DocGraph is a lens over the plain HTML and Markdown on your disk. Close the app and everything is exactly where you left it. [How privacy works →](https://docgraph.app/privacy) - **Local-first** — plain files on your machine, always readable. - **No telemetry** — none at all — the app never reports on you. - **AI runs locally** — a built-in model, Ollama or Apple's — cloud only if you add a key or connect an outside agent. - **Reads your formats** — PDF, Excel, CSV and Word — no import. - **Free to use** — no account, no payment, no tier. ## Your knowledge. Your machine. Your graph. See how your knowledge connects. [Get DocGraph — free](https://docgraph.app/download)