Forsion

Remember it. Think it through. Keep it moving.
A local-first AI second brain you own.

English · 简体中文

!Platforms

Download · Quick start · Documentation (中文) · Website

Forsion is a local-first AI second brain. Organize your own Markdown files in a Notion-style block editor, think with Agents that retain useful context, build a TEAM to move projects forward, and let Muse follow up within the boundaries you set.

Make each piece of thinking a starting point for the next. Sources, preferences, discussions, outputs, and follow-ups come together in one workspace.

| Part of your second brain | How it works in Forsion | | --- | --- | | Capture and connect | Block editing on local .md files, linked to notes, PDFs, whiteboards, databases, and conversations. | | Remember context | Inspectable, editable long-term memory that carries preferences and project background across conversations. | | Think and act together | Agents with distinct personalities, persistent TEAMs, parallel work, and Team Desk. | | Follow through | Muse wakes on heartbeats, schedules, and rules, bringing results and decisions to Inbox. | | Extend your workflow | Plugins bring tools, Agents, skills, and dedicated workspaces into the same flow. |

The screenshots below come from the v2.11.0 desktop build and use one public research-project fixture. They show the real interface, not a live model run. Capture provenance and reproduction.

Markdown files. A block editor you can think in.

Notion-style block editing, built on local .md files. In the Amadeus knowledge base, headings, paragraphs, to-dos, code, and embedded content become blocks you can organize. Write with Markdown shortcuts, insert content with /, arrange sources in columns, and inspect block source. Your notes remain files on disk that you can open, back up, and ask an Agent to read or update.

_{Insert blocks through the slash menu in a local Markdown note · v2.11.0 interface · Public sample data}

• Give a page structure. Markdown shortcuts, slash commands, columns, section folding, callouts, page icons, and covers help organize longer notes.
• Connect your material. Use [[wikilinks]], backlinks, and a graph to connect ideas. Embed notes, images, databases, and whiteboards; reference notes directly in chat.
• Put sources to work. Within the access you authorize, Agents can read sources, compare approaches, and write conclusions and next steps back to the project. You and your Agents work with the same actual files.

Notes use local Markdown, with compatible formats for links, properties, and whiteboards. Rendering of extensions such as columns, databases, and plugin blocks depends on the application opening them.

Block editor (中文) · Knowledge base (中文) · Databases (中文) · Whiteboards (中文)

Keep the background worth carrying forward

Your knowledge base holds the material. Memory carries useful context. Full articles, evidence, and outputs stay in the project. Preferences, facts, and lasting background you explicitly share can become Agent memory, giving the next conversation somewhere to begin.

• See what it remembers. Inspect, revise, and organize memories in the memory panel. Check sources and revision history, or remove an item from active memory.
• Give each Agent its own continuity. Agents have a personality (SOUL), memory, Library, and work journal, with individually manageable models, tool permissions, and memory settings.
• Maintain the accumulated context. Historian organizes logs and memory candidates. Enable Dream when you want memory consolidation; conversation recall searches relevant history within the Agent's scope.

_{Inspect project context and preferences in Arioso’s memory panel, with edit and forget controls · v2.11.0 interface · Public sample data}

Memory (中文) · Configure Agents (中文)

TEAM: room for imagination, scrutiny, and shared work

A solo project can still have different perspectives and lasting roles. Aria attends to feeling and creative possibilities; Recita examines evidence, assumptions, and feasibility; Arioso weighs your aims and tradeoffs with independent judgment. Coding focuses on implementation, while Muse follows up proactively. Customize them or create your own Agents.

TEAM in v2.11 makes collaboration something you can build on:

• Shared expectations. Select two or more Agents to start team mode. Persistent teams retain members, roles, and a custom avatar, have a Library, and use TEAM.md for common goals and working agreements.
• Parallel work. Members research, write, or execute in their own child conversations, posting progress, questions, and handoffs to the main conversation. Interject or @ a member to keep the work focused.
• Visible progress and outputs. Team Desk shows who is working, awaiting approval, or done. Open a member's full child conversation; review approval requests and public outputs in the main conversation.
• A home for each task. Use @project in a private chat to dispatch work into a new conversation in that project. Agents Space manages Agents and teams; the Orbits sidebar brings private chats, teams, projects, and external engines together.

_{Public discussion from Aria, Recita, and Arioso alongside member status in Team Desk · v2.11.0 interface · Public sample data}

_{Open Recita’s child conversation to examine evidence and suggestions beside the main discussion · v2.11.0 interface · Public sample data}

TEAM and Team Desk · Agents and personalities (中文) · External engines (中文)

Muse: keep the follow-up moving

Hand ongoing work to a proactive Agent. Once enabled, Muse wakes on heartbeats, schedules, and rules to review recent work, organize material, suggest next steps, or carry out tasks within configured permissions and budgets.

• Hand over a next step. Assign a conversation task card to Muse, or configure schedules and event rules. Heartbeats let it periodically revisit work worth moving forward.
• Bring decisions back to you. Results arrive in Inbox. Execute a task with Muse, continue in a new conversation, or dismiss it; approve or reject pending actions.
• Give ongoing work a place. Muse builds its own Library and journals, and can write and gradually improve its own Space to make that work browsable.

_{Review a Muse follow-up in Inbox, inspect the task, execute it, start a new conversation, or dismiss it · v2.11.0 interface · Public sample data}

Proactive Muse (中文) · Automation (中文) · Inbox (中文)

Plugins: bring your way of working

Your second brain can grow with your work. A plugin bundle can deliver an interface, engine tools, Agents, skills, and a Space together, connecting collection, processing, and output.

• Collect and act. Bluebird turns videos into notes; Computer Use lets Agents operate desktop applications on supported platforms; plugin events can trigger automation.
• Reuse a method. Skills preserve ways of doing tasks, MCP connects external tools and data, and plugins extend note blocks, views, and workspaces.
• Build your own environment. Install plugins, Agents, skills, Spaces, themes, and web apps from the store. Agents can also use built-in extension-development skills to create extensions for you.

From a few sources to a research talk you can keep developing

Try building up this workflow with a real project:

1. Create a project note. Organize goals, sources, and questions with blocks; embed PDFs, whiteboards, or databases.
2. Think with context. Reference your material and a confirmed writing preference. Ask Aria to explore directions, Recita to examine evidence, and Arioso to help weigh the options.
3. Divide the work with TEAM. Assign independent investigations, drafts, or checks to members, then follow progress and outputs in Team Desk.
4. Write decisions back to files. Review the result and ask an Agent to save the outline, sources, and open questions. Inspect and edit them in Agent Desk; continue in Coding Studio if you need a webpage prototype.
5. Leave a follow-up. Ask Muse to revisit the project at an agreed time, handle its results in Inbox, and add new conclusions to the project.

_{Keep sources, team responsibilities, and next steps together in a local project note · v2.11.0 interface · Public sample data}

Work in an environment you control

• Arrange your thinking space. Spaces, split panels, draggable tabs, independent windows, and Mini Panel keep conversations, material, and outputs side by side, with restorable layouts.
• Talk through the work. Hands-free live voice sends after you pause; the call continues when you switch to other views.
• Choose your models. Use Forsion-hosted models, your own API connection, OpenAI-compatible endpoints, local Ollama, or supported subscription logins.
• Reach Agents elsewhere. Configure WeChat, Telegram, or QQ channels. Web and Android offer cloud-connected access, with capabilities that differ from Desktop.
• Keep access to your accumulated work. Inspect and back up local files, and edit memory. Cloud models, account sync, and sharing send the relevant content to the services involved.

Download and start

Download the latest desktop installer from GitHub Releases.

| Platform | Release artifact | Notes | | --- | --- | --- | | macOS | Forsion-*.dmg | Apple Silicon (arm64). | | Windows | Forsion-*.exe | NSIS installer. | | Linux | Forsion-*.AppImage | Add execute permission, then run. |

Every desktop release includes two distributions. Forsion-- bundles Forsion Extend for account and cloud features. Forsion-NoExtend-- includes the local app and Agent backend without bundling Extend; cloud account, sync, publishing and remote device features require Extend. Both use the same app identity and data folders. Windows/Linux updates stay within the chosen distribution; on macOS, download the same filename variant when updating.

For local builds, run FORSION_BUNDLE_EXTEND=0 npm run build, then FORSION_BUNDLE_EXTEND=0 npm run dist from desktop/. Omitting the variable produces the default distribution.

1. Install and connect a model. First launch helps configure your connection, model, and workspace. The desktop installer includes the Agent backend and Node.js runtime.
2. Bring some real context. Create a project note with a goal, sources, and next steps. Ask an Agent to read it, and explicitly tell it a preference worth remembering.
3. Finish one useful task. Ask an Agent to produce something from your material, inspect the saved file, and check an explicitly saved preference in the memory panel. Add TEAM, Muse, and plugins when you need them.

Installer signing: macOS builds use ad-hoc signing and are not yet notarized; Windows builds may trigger SmartScreen. Download from this repository's Releases. If Gatekeeper blocks the first open on macOS, right-click the app and choose “Open”, or allow it in System Settings → Privacy & Security.

Where Your Data Lives

The official desktop build uses, by default:

| Path | Contents | | --- | --- | | ~/.forsion/ | Account, settings, Agent data, sessions, skills, plugins and local databases. | | ~/Forsion/ | User-visible workspace, knowledge base and project files. |

The dev build uses separate ~/.forsion-dev/ and ~/Forsion-Dev/ so it won't pollute production data. Legacy ~/.tangu / ~/Tangu data is migrated by the desktop app, which keeps a compatibility entry point.

Back up these two directories regularly, just like ordinary documents. Before running high-privilege Agent tasks, confirm the approval tier and the current working directory.

For developers

Forsion Genesis is the source repository for the product family: the Tangu Agent runtime, the LCL Space / View / Plugin engine, Desktop, Web, Android, and product profiles such as Amadeus.

See web/README.md for Web deployment and mobile/README.md for Android builds. Expand below for all development commands and the architecture.

git clone https://github.com/Changan-Su/Forsion.git
cd Forsion
Build the Agent backend embedded in the desktop app first
cd tangu-agent
npm ci
npm run build
Then start the Electron desktop app
cd ../desktop
npm ci
npm run dev

cd tangu-agent
npm ci
npm run build
Optional: copy and edit local config
mkdir -p ~/.tangu
cp example.env ~/.tangu/.env
npm run tui       # Terminal interface
npm run server    # HTTP / SSE service, default 127.0.0.1:8787

cd desktop
npm run dev:amadeus
npm run build:amadeus
npm run pack:amadeus

cd web
cp .env.example .env
npm ci
npm run dev

BACKEND_URL=http://host.docker.internal:3001 \
  docker compose -f web/docker-compose.yml up -d --build

flowchart LR
    Agent["Tangu Agent Runtime"] --> TUI["Terminal TUI"]
    Agent --> Service["Standalone HTTP / SSE"]
    Agent --> Desktop["Forsion Desktop"]
LCL["LCL Space / View Engine"] --> Desktop
    Renderer["Shared React Renderer"] --> Desktop
    LCL --> Web["Forsion Web"]
    Renderer --> Web
    LCL --> Mobile["Android Client"]
    Renderer --> Mobile
Web --> Gateway["Forsion Server"]
    Mobile --> Gateway
    Gateway --> Worker["Tangu Worker"]

Forsion/
├── tangu-agent/   # Agent runtime, TUI, standalone, tools, skills & plugin API
├── lcl/           # Space / View / Plugin workspace engine
├── desktop/       # Electron main process, shared React renderer & product profiles
├── web/           # Vite + nginx browser client
├── mobile/        # Capacitor Android client
├── unit/          # Standalone Forsion Unit framework & local plugins
├── docs/          # User documentation
├── scripts/       # Open-source boundary check
├── .github/       # CI workflows
├── archived/      # Read-only historical implementations
└── Dockerfile.standalone  # Worker image used by Tangu Manager

cd tangu-agent
npm run typecheck
npm test
cd ../desktop
npm run typecheck
npm test

Contributing

Issues and Pull Requests are welcome. To make problems easier to reproduce and merge:

1. When filing a bug, include the Forsion version, operating system, reproduction steps, expected behavior and any necessary logs;
2. Search existing issues before development, and keep each Pull Request focused on one clear problem;
3. Do not commit .env, tokens, local databases, workspace content or build artifacts;
4. Behavior changes need added or updated tests, and user-visible changes need a Changelog update;
5. When modifying the shared rendering layer, consider the capability differences and runtime gating across Desktop, Web and Mobile.

License

Forsion Genesis uses the Modified Apache License 2.0. It is based on Apache License 2.0 with additional conditions, mainly covering:

• You may not use this repository's code to operate a multi-tenant environment without written authorization;
• When using the desktop/, web/ or mobile/ frontends, you may not remove or modify the logo, product name and copyright notices in the interface;
• Commercial use is permitted when the additional conditions are met; a commercial license should be obtained for multi-tenant deployment or other authorizations.