> ## Documentation Index
> Fetch the complete documentation index at: https://docs.taskbean.ai/llms.txt
> Use this file to discover all available pages before exploring further.

> ## Agent Instructions
> Treat taskbean/taskbean as the source of truth for product behavior.
> Use the canonical Project, Project ID, Project Key, Project Alias, Project Selector, Workspace, Tracked Project, Tracked Workspace, Discovered Workspace, Remote Workspace, Primary Workspace, Agent Session, Copilot Fork, Task, Unassigned Work, Execution Profile, Kickoff Prompt, Execution Record, and Execution Step terminology from the glossary.
> Execution terms are domain definitions, not evidence of available CLI commands or desktop controls.
> Do not infer that Windows-specific desktop features apply to the cross-platform CLI.

# Get started with the desktop app

> Launch the Taskbean desktop PWA with Portless, Python, or the Windows launcher and check backend readiness.

The desktop app is the human-facing half of Taskbean. It shares `~/.taskbean/taskbean.db` with the CLI and adds on-device AI, task review, reminders, agent usage, Chronicle review, and telemetry.

<Warning>
  The current app dependency set is Windows-oriented. The `bean` CLI remains the cross-platform surface.
</Warning>

## Prerequisites

* Windows 10 or 11
* Python 3.10 or later
* [Foundry Local](https://github.com/microsoft/Foundry-Local)
* Node.js 24 or later only when you use the Portless path
* Docker only when you want the optional Jaeger trace view

## Start with Portless

Portless provides a stable local HTTPS URL:

```bash theme={"system"}
git clone https://github.com/taskbean/taskbean.git
cd taskbean/app
npm install
npm run dev
```

Open `https://taskbean.localhost`. Portless runs `python agent/main.py` through the `dev:agent` script.

## Start Python directly

```bash theme={"system"}
cd taskbean/app/agent
pip install -r requirements.txt
python main.py
```

Open `http://127.0.0.1:8275`.

## Use the Windows launcher

From `taskbean/app`:

```powershell theme={"system"}
.\launch.ps1
```

You can also run `launch.cmd`. It delegates to the PowerShell launcher.

## Check readiness

| Endpoint | Meaning |
| - | - |
| `GET /api/health` | The process is alive. It always returns `200`; inspect `modelReady` and `startupError`. |
| `GET /api/ready` | The model is ready. It returns `503` while loading, `500` on startup failure, and `200` when ready. |

The first model load can take up to several minutes on some NPU paths.

## Install the PWA

Open the app in Edge or Chrome, then use the browser install control. The installed shell opens in a standalone window. Cached UI assets and selected read-only API responses can load without internet access, but live task data and AI inference still require the local Python backend.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.