# Wavefunction

> Open-source desktop coding agent for macOS that runs parallel chats and subagents with local models, API keys or a ChatGPT sign-in.

Wavefunction is a free, MIT-licensed desktop coding agent for macOS. It centers on a single conversation with one built-in agent that can work on several tasks at once through parallel chats and subagents. You can use local models, your own API keys or a ChatGPT sign-in.

## What It Is

Wavefunction is a minimal desktop app for working with a coding agent. Projects are folders, usually Git repositories, and each chat is a separate conversation with its own model, folder and history. The interface keeps the conversation at the center, with a project sidebar and a composer. It has no built-in browser, and links open in your default browser.

## How Parallel Work Happens

You can send a new task while earlier ones are still running. For bigger tasks the agent starts helpers, which are child chats that work in parallel and report back. In a Git repository a new chat starts on its own branch in its own checkout, and helpers that might touch the same files can get isolated worktrees. The parent reviews their changes and brings them together. A message sent mid-task steers the running task, or can be queued as a follow-up. You can fork a chat from any message or rewind the code and conversation to an earlier point. Plan mode reads and explains without changing anything, and routines send the same instruction to a chat on a schedule while the app is open.

## Computer Use in a Container

When the agent needs a desktop, it uses its own, running in a local Docker container. It does not move your pointer, type on your keyboard or record your screen. Sign-ins are done privately, so the agent never sees what you type. Files move in and out through folders you choose. This feature requires Docker Desktop and a model that can read images.

## Models, Connectors and Privacy

Supported providers include ChatGPT sign-in, OpenAI, Anthropic, Google Gemini, OpenRouter, Vercel AI Gateway, Ollama and OpenAI-compatible servers such as LM Studio. The app offers 70+ ready-made MCP connectors plus any MCP server you add. It follows AGENTS.md instructions and loads SKILL.md skills from supported folders. Conversations and settings are stored locally, and a zero-data-retention option applies to OpenRouter and Vercel AI Gateway. Commands run in the macOS sandbox by default, with an ask-first check for commands it recognizes as irreversible. A usage analytics view tracks provider-reported tokens, cache usage and costs.

## Setup Path

Download the signed and notarized DMG from the latest release. It requires a Mac with Apple Silicon running macOS 11 or later. Setup walks through connecting a provider, choosing models and creating a first project. The app can also be built from source. It uses a Tauri shell in Rust, a React interface and a Python agent engine.

## Features
- Parallel chats and subagents
- Per-chat git branches and worktrees
- Steer or queue follow-ups during a task
- Fork and rewind chats
- Plan mode and full access mode
- Scheduled routines
- Computer use in a local Docker container
- 70+ MCP connectors plus custom MCP servers
- AGENTS.md and SKILL.md support
- Bring your own model with local, API key or ChatGPT sign-in
- Zero data retention option
- Token and cost analytics
- Automatic context summarization
- macOS sandboxed command execution

## Integrations
ChatGPT, OpenAI, Anthropic, Google Gemini, OpenRouter, Vercel AI Gateway, Ollama, LM Studio, Docker Desktop, MCP, Gmail, Outlook, Microsoft 365, Airtable, Asana, Atlassian

## Platforms
MACOS, IOS, WEB, API, CLI

## Pricing
Open Source

## Version
0.4.0

## Links
- Website: https://wavefunction.si
- Documentation: https://github.com/stoyan-stoyanov/wavefunction/blob/main/docs/DEVELOPMENT.md
- Repository: https://github.com/stoyan-stoyanov/wavefunction
- EveryDev.ai: https://www.everydev.ai/tools/wavefunction
