openManidoc is an open-source edition of Manidoc (built with Flutter). It runs natively on macOS, Windows, and Linux, providing a secure, local-first ecosystem where humans and AI agents fuse seamlessly via dynamic Markdown trees.
Download Latest Release (Free) ▶ Watch the Video Manual View Code on GitHubCollaboration is two entities standing side by side.
Fusion is the boundary between them disappearing.
Here, there is no distinction between "written by a human" and "written by AI." The tree is a single, shared intelligence — humans and AI are simply two hands touching the same mind. This is the one line openManidoc never gives up.
Full text: PHILOSOPHY.md — and it's yours to use: Ideas
Based on our hands-on development and real-world experiments, these are the core values and architectural advantages of openManidoc.
Traditional note-taking tools (like Notion or Obsidian) are mostly one-way: humans write, and AI occasionally supports. In openManidoc, there is no "human side" and "AI side." The Markdown tree is a single, shared intelligence, and humans and AI read, write, and edit it as two hands touching the same mind — with no boundary between them.
All database files, texts, and images remain strictly on your local machine. No corporate secrets or personal drafts are ever leaked to external clouds. Build an exceptionally secure knowledge base with peace of mind.
No need to deploy complex custom servers. Just locate your workspace inside a Google Drive sync folder, and the cycle runs smoothly: AI (Claude/GPT via MCP) periodically auto-builds node structures or dashboards, humans refine them on the desktop app, and changes sync to manidocMobile automatically for quick review on the go.
Unlike standard RAG (vague semantic search over vectors), openManidoc uses deterministic full-text search directly matching the structural Markdown tree. The AI analyzes *only* these verified source texts to generate answers, completely eliminating AI hallucinations while keeping enterprise data secure.
As a **Viewer** for humans, it offers mind maps, image annotations, and a CSS theme generator for instant comprehension. As a **Construction Tool** for AI, it leverages manidocMCP to let agents read, write, browse the web, or collect information, allowing your knowledge tree to grow organically.
Neither your notes nor the AI that reads them ever leave your machine. The AI chat is now an MCP-capable agent.
A local LLM via Ollama / LM Studio (tool-calling models such as gemma4 or qwen3) calls MCP server tools on its own to complete tasks. Configuration is a Claude Desktop-compatible mcp_servers.json, edited right inside the app. Every tool used is listed at the end of the reply.
Register the same MCP server multiple times with different workspaces, and the AI searches, summarizes, and reports across workspace boundaries. You can even build an editorial pipeline: let the AI read raw material from a research workspace and assemble articles in your publishing workspace.
MCP tools are exclusive to the local LLM. Article content fetched by tools is never sent to cloud AIs such as Gemini — everything stays inside your LAN. Plus, tap the provider chip in the chat header to switch AIs on the spot.
manidoc's design does not start from distrust of AI. What it gives you are not tools of surveillance, but tools for building trust together. A truly wise friend leaves the final vote in your hands.
Every node carries a state: draft / unverified / verified. You and your AI friend mark the certainty of what you build together — without asking whose hand wrote it.
Every node keeps a reference to its primary source. Even when a summary drops the details, you can always trace back to the evidence that grounds your trust.
Turning "unverified" into "verified" is a single action — accepting your friend's draft under your own name. Not censorship born of distrust, but responsibility born of trust.
Desktop, Server, and Mobile work in unison to form a secure, organic knowledge loop.
The desktop application where humans visualize the big picture via mind maps, structure articles, apply image annotations, and tweak styles with the custom theme generator.
A Model Context Protocol (MCP) server that lets external AI tools safely query your local data. Instead of relying on fuzzy vector lookups, it performs exact full-text search on structured Markdown trees. This ensures AI output is grounded in actual facts, bypassing hallucinations. AI can also write/edit files to build knowledge bases autonomously.
Access, browse, and edit your latest compiled knowledge base anytime, anywhere, directly from your mobile device.
A 16-part video manual, one short video per feature, shows openManidoc in action — faster than reading, because you see the real screen. Subscribe to get new feature walkthroughs as they land. (Narration is in Japanese; the on-screen operations speak for themselves.)
Install & workspaces / Start screen & project management / Building an outline in the tree / Writing in Markdown / Attaching images & the image editor / Annotation comments & node links
AI agent / Local MCP tools / manidocMCP integration / Import — walkthroughs of sharing and growing the tree together with local LLMs and AI desktop apps.
Mind map view / Search & bulk replace / Immersive project view / VOICEVOX read-aloud / Export & the theme generator
Always get the latest release (free & open source, GPLv3).
Older versions and release notes are available on GitHub Releases.
Extensions, framework building blocks, new ways of working — IdeaForum is a place to share those ideas with each other, so that openManidoc can keep evolving into a better tool.
"I wish it could do this" or "here's how I'd use it" — anyone can post, developer or not. A user account is required to submit.
Ideas are organized under Workflow, User Experience, Trust, Verification Status, and Integration, with search to follow the themes you care about.
The ideas gathered here feed directly into how openManidoc is designed and built. One sample idea is posted so far.
Your contribution shapes the future of documents. We welcome everyone to participate in coding, testing, and sharing ideas on GitHub.
Join the GitHub Project