MultiAgentOS
Donate & download

Native Mac & Windows

A desktop AI helper that gets things done for you.

Chat with a friendly AI assistant that can open a browser, control apps, run commands, and work with your files, all in one window on your Mac or PC. Use popular cloud AI like OpenAI and Anthropic, or run private AI right on your own machine. You see everything it does and approve anything sensitive.

Donate & download Read the full guide See the UI
macOS 14+ Windows 10+ Always on top Resizes with context Works offline Your data stays local
MultiAgentOS desktop app: a left navigator with bookmarks, history, and downloads; the browser workspace in the centre; and the Copilot Agent Inspector with Chat, Plan, Activity, Artifacts, and Context on the right.
The MAOS workspace: a built-in browser the agent drives, with the Bridge chat panel docked below so you see every step.

The right tool for the job built in

Ask it to edit code, browse the web, run a command, or use a desktop, and the matching workspace opens right inside the app.

Learn how it works
  1. 25everyday actions ready to use
  2. Workspaceswatch the work happen
  3. 5 packsextra tools loaded on demand
  4. Helpershand off a task with limits you set

One tidy window

Everything happens in one place.

No juggling separate apps. Your past chats sit on the left, your conversation in the middle, and whatever the assistant is working on, editing a file, running a command, or handing a task to a helper, opens neatly on the right. It all stays in view.

1

Edit files

Open a file or folder, let the assistant make changes, and save the result, right beside your chat.

2

Task helpers

Hand off a job to a focused helper with a clear goal and limits you set, then see the results come back.

3

Connect your AI

Set up your AI providers and on-device models in one panel, then test each one with a single click.

4

Run commands

Type or let the assistant run commands in a built-in terminal, with everything visible as it happens.

MultiAgentOS in-app browser showing a travel-deals article the agent navigated to, with the MAOS Bridge chat below showing the user's request and the agent's plan.
Browse on the pageAsk the agent to look something up and watch it open and read the actual source page, right in the window.
MultiAgentOS Settings on the LLM Connection API Key tab with a provider set to DeepSeek, a model name, a masked API token, and reasoning controls.
Bring your own AIConnect any provider with your own API key and pick the exact model you want to run.
MultiAgentOS Settings on the LLM Connection Local Server tab with a localhost endpoint, the llama3 model, reasoning level, system prompt, and MCP servers JSON field.
Run it locallyPoint MAOS at your own Ollama, LM Studio, or llama.cpp server so the model runs on your machine.
MultiAgentOS Settings showing the MCP Servers JSON editor with helper text on tool naming, a status line, and Connect, Disconnect, and Test buttons.
Add your own toolsWire in MCP servers and a system prompt, then test the connection before you commit.

Watch it work

It does the task, and you see every step.

Ask MultiAgentOS to handle a web task or use a desktop, and you watch it happen in real time. It opens a browser, fills forms, clicks around, or works on its own in-app desktop, so nothing is hidden and you can step in anytime.

01

Ask for what you want

Tell it the task in plain words and set what it is allowed to do before it starts.

  • Your whole workspace stays in view
  • Pick which AI to use right from the chat bar
  • You approve anything sensitive before it runs
02

Let it browse the web

It opens a real browser, reads the page, clicks, types, and tells you how it is going.

  • Opens right inside the app
  • Reads, clicks, types, and keeps you posted
  • Stop or review at any moment
03

Give it its own desktop

It works in a contained desktop inside the app, so your real desktop stays untouched.

  • Stays in its own safe space
  • Keeps the task right next to your chat
  • Your own desktop stays separate and private

Bring your own AI, six easy ways.

Use the popular cloud services, or keep everything private on your own computer.

Cloud AI (API key)

Paste a key from OpenAI, Anthropic, DeepSeek, xAI, Google, Mistral, and more.

Local Server

Connect AI running on your own machine through Ollama, LM Studio, GPT4All, and similar.

Your own command

Already have an AI command line tool? Plug it in and MultiAgentOS uses its answers.

Sign-in (OAuth)

Sign in securely to Google Vertex AI, Azure OpenAI, HuggingFace, and more.

Terminal

Route your questions through a saved terminal command and get the reply back automatically.

Private on-device AI

Point it at an AI model file on your computer and run completely offline, no internet needed.

How to set it up and use it

From install to your first agent task.

MultiAgentOS is ready to work in minutes. Here is the whole path: install it, connect the AI you want (cloud or fully private), then ask in plain words and watch the agent browse, run tools, and hand back answers you can use.

01

Install and open

Install MultiAgentOS on your Mac or PC and open it. You land in the workspace: a built-in browser with the MAOS Bridge chat panel docked below.

  • Native app for macOS and Windows
  • No accounts to create
  • Your data stays on your machine
02

Connect your AI

Open Settings → LLM Connection and pick a tab. Use an API key for a cloud provider (OpenAI, Anthropic, DeepSeek, and more), then press Connect and Test.

  • Dozens of providers built in, or a custom endpoint
  • Keys are stored in the system keychain
  • Set a reasoning level and system prompt
03

Or keep it private

Prefer to stay offline? Point MAOS at a Local Server (Ollama, LM Studio, llama.cpp), or load a Local AI GGUF model file, so the model runs entirely on your machine.

  • Local Server at an OpenAI-compatible URL
  • Built-in local model file for full offline use
  • CLI pipe and terminal options for power users
04

Add your own tools

Paste MCP server definitions as JSON and set a system prompt to give the agent extra abilities and data. Tools appear to the model as mcp__name__tool.

  • Connect MCP servers under your control
  • Test before you commit
  • 25 everyday actions ready out of the box
05

Ask in the Bridge

Type a request into the Bridge chat, like “search the browser for current China travel deals.” The agent navigates, reads the page, and runs tools, showing how many tool runs each step took.

  • Watch it browse live in the same window
  • See a tool-run count under each reply
  • Attach or paste images for vision
06

Get answers you can use

The agent distils its browsing into clean tables, ranked recommendations, and summaries, right in the chat. You approve anything sensitive before it runs.

  • Markdown tables and shortlists
  • Copy or refine with a follow-up
  • Approval walls keep you in control
MultiAgentOS Settings LLM Connection page with the provider dropdown open, listing many supported providers such as OpenAI, Anthropic, Groq, OpenRouter, DeepSeek, Together, Mistral, and a Custom option.
Pick any providerChoose from dozens of built-in AI providers, or add a custom endpoint of your own.
Close-up of the MAOS Bridge chat showing a completed step labelled 3 tool runs, a user request to search the browser, the agent's plan, and 16 tool runs, above the message input box.
See every tool runEach reply shows how many tool runs it took, so you always know what the agent actually did.
MAOS Bridge chat showing an agent-generated Markdown table of China tour packages with columns for tour, duration, discount, and per-person price.
Answers you can useGet the agent's browsing distilled into a clean comparison table you can read at a glance.
MAOS Bridge chat showing an agent reply headed Best Value Picks with a numbered list of recommended trips and prices.
Ranked recommendationsLet the agent weigh the options and hand back ranked best-value picks tailored to your ask.

Everything you need, right where you work.

One simple window. Lots of ways to get things done.

+ Attach

Inputs

Text, files, folders, screenshots, or your voice.

Window sizes

Pill, half-screen, tall pane, or full command center.

Run commandOpen folderScroll page

Real actions

It can run a command, open a folder, scroll a page, and more, on your say-so.

report.pdf

Drag & drop

Drop files into the app to attach context instantly.

Command templatesollama run qwen2.5:7b "{prompt}"

Saved commands

Save the commands you use often and pick one in a click.

Add more skills

Plug in extra tools and data sources to teach it new tricks (via MCP).

Settings that stay out of your way.

Configure once. It just works.

One settings sheet

All connections, models, prompts, and servers in one place.

Pick the right model

Choose which AI handles each task, or let MultiAgentOS decide for you.

Set the ground rules

Give standing instructions that shape how it answers in every chat.

Test connections

One-click connection tests for every provider and server.

MultiAgentOS Settings Done

All good! Your connections are ready.

Built to work well, big AI or small.

Out of the box it has 25 ready-to-use actions for clicking, typing, files, the web, running commands, and handing tasks to helpers. Extra tools stay tucked away and load only when a job needs them, so even smaller, private AI models running on your own computer stay quick and focused instead of overwhelmed.

25Actions ready to go 5Extra tool packs 8Max steps per helper

Support the project and download MultiAgentOS.

Donate to support the build and download a real installer right away — each download link is single-use.

Donate & download View all features

Support the project $10

per month

All listed software included
  • Access to all listed software products
  • All software is currently in development
  • Launch links are sent by email
  • Books remain separate one-time purchases

Get the software

Support the project and download a build.

Donate what you like, choose a product and your platform, and download a real installer right after checkout. Each link is single-use and time-limited.