What it does

Coworker is a desktop app for working with an AI coding agent on your own projects. It pairs an Electron interface with a local backend that stores memory and chat logs on your computer. Prompts go to whichever model provider you configure, so the work stays fully on your machine only when you use a local model such as Ollama. You choose the model. It ships with 32 provider presets, accepts any OpenAI-compatible endpoint, and can talk to local models through Ollama. Provider keys are stored in the macOS Keychain or a permission-restricted file. The agent can search the web, browse, manage files, run commands in an integrated terminal, hand work to sub-agents, and call MCP servers or installable skills. Commands and file changes wait for your approval by default, and every change is logged with before and after diffs. The project calls itself a fast-moving pre-release. Multi-agent teams are experimental, macOS is the most polished platform, Windows has some beta features, and Linux has barely been tested.

Capabilities

Local-first: memory and chat logs are stored on your machine, and local Ollama models are supported
Works with OpenAI-compatible APIs, Ollama and custom endpoints
Approval step for commands and file changes by default
MCP integration plus a SKILL.md skills system with marketplaces
Long-term memory, goal mode and an integrated terminal
Audit trails, agent trace export and change tracking

When to use it

You want an AI coding assistant that stores its data locally and can run on local Ollama models
You want to switch between cloud models and local Ollama models
You want to review every command and file change the agent proposes
You are comfortable running a pre-release app with rough edges