Coworker Developer Guide
Coworker is an open-source desktop AI agent that runs locally and helps with files, documents, browser tasks, and repeatable workflows. This guide explains how Coworker is positioned, how the desktop setup works, how its local-first model shapes architecture and safety, and where to contribute.
This guide is based on the official accomplish-ai/coworker repository, checked on June 26, 2026.
In this section​
- Setup and first run
- Architecture and runtime
- Configuration and security
- Extending and contributing
- Simple user guide
1. What Coworker is​
The shortest accurate description is:
Coworker is a local desktop AI agent that acts on files, documents, browser tasks, and connected tools while letting you bring your own model provider or local model runtime.
The README emphasizes:
- local execution,
- your own API keys or local models,
- file management,
- document creation,
- browser tasks,
- and user approval for actions.
2. The mental model​
Think of Coworker as five cooperating layers:
| Layer | What it does |
|---|---|
| desktop app | Main user interaction surface |
| local action runtime | Works on files, documents, and browser flows |
| provider layer | Connects cloud or local models |
| tool connections | Reach systems like Notion or Drive |
| review and control layer | Lets the user approve and inspect actions |
3. Why developers would choose it​
- It is clearly local-first.
- It supports both API providers and local models.
- It is aimed at practical desktop work, not only chat.
- It presents itself as a tool, not a service, which many teams will appreciate.
4. What makes it distinctive​
Coworker tries to sit between a plain chat app and a heavy agent platform. It focuses on real local productivity work while keeping the user in control of access and review.
5. Recommended reading order​
- this overview,
- setup and downloads,
- architecture and runtime,
- configuration and security,
- extension and contribution surfaces.