Open-source AI workspace
Plan with one model, build with another, and keep the whole context — on your computer. When the work should keep going without you, hand it to a bot with a computer of its own.
macOS, Windows and Linux · MIT license · Bring your own accounts
Real app · demo data
Two ways to work
Combine models and accounts in one conversation, let Maestro delegate to a team of agents, and keep code, terminal and browser beside the chat.
Bots with their own Linux desktop, on your own VPS or this computer's Docker. They keep working while your laptop is closed, and ask you when it matters.
Connect what you already use
Model availability and limits follow each provider and plan. Compare providers
On your computer
Connect the subscriptions and API keys you already have — ChatGPT, Claude, GitHub Copilot, Cursor, Google AI, Grok — and use them together.
Plan with Claude Fable, build with GPT Sol, review with Opus. History, plan and tool results stay in the same conversation when the next model takes over.
Connect several ChatGPT or Claude accounts. With automatic rotation, the next account picks up when one hits its limit — conversation and finished tool results included.
An orchestrator splits the work across subagent profiles — explorer, frontend, tests, review — each with its own model, even from different providers. Independent tasks run in parallel.
Open VS Code, terminals and an embedded browser beside the conversation. Each conversation keeps its own tools, and Git worktrees keep parallel work apart.
Agents publish prototypes, reports and dashboards as versioned artifacts. Send a link and collect comments on the exact passage — your agent can read and answer them.
Always onExperimental
Run Maestrly bots on a server you control. Each one has a Linux desktop with a browser and terminal, its own memory and routines, and the model accounts you give it.
Follow a bot's conversation beside its live screen. Take control with your own mouse and keyboard, leave it a note, and give it back.
Permission requests and questions from every bot land in Awaiting you, with an alert when one finishes or needs you. A bot can never approve for you.
Give a bot a prompt and a schedule — daily, on chosen days, or every few hours. Routines run on the server while your computer is off, each run in a conversation of its own that never holds up yours.
Each of your computers, or a workspaces environment of your server, decides which projects, models and actions a bot may use there. Its conversations run in their own worktrees, and plan approvals always stay with you.
Install the bot server on this computer's Docker or on an Ubuntu or Debian VPS over SSH. No certificates or VPN — Maestrly pairs it, and keeps it updated.
And a lot more
Start without a project, with private files and permissions, and send the plan to a project later.
Personal and project memories, recalled for each message. Automatic saving is opt-in.
Speak to a chat or a bot; transcribed on your computer with whisper.cpp, or by OpenAI or Groq with your key.
Ask for a navigable prototype with simulated data before you build the real thing.
Install Codex, Claude Code and Agent Plugins plugins from Git. Their skills, MCP tools, agents and hooks work with every model.
Bring ChatGPT web to your project, with scoped tools and plans sent to your agent.
Free · Open source · Local-first
The whole app is open source. Contributions are optional and never unlock features.
No sign-up, analytics or telemetry. Conversations and project state stay on your computer.
Prompts go straight to the providers you choose. Bots run only on infrastructure you control.
Before you start
Yes. The app is free and open source under the MIT license. You bring your own accounts or API keys, so provider costs and limits still apply. Contributions are optional.
No. The desktop app works on its own. The bot server is optional: you need it for bots, for hosting shared artifacts, and to run your chats and workspaces away from your computer. You run it yourself — on this computer's Docker or your own VPS.
Yes, with a bot server on a Linux VPS (experimental). Bots get their own desktop, browser, terminal and model accounts, and their routines keep running. You can also run your own chats and workspaces on that server: a turn keeps going while your laptop is closed, and the app shows what happened when you come back. Bots can work in those workspaces too.
Yes. Switch providers within a conversation and give Maestro subagents from other providers. Available models and features depend on the accounts you connect and the permissions you grant.
Conversations, settings and project state stay on your device. Your chosen providers receive the context needed for each task, and tools follow your permissions. Bots, shared artifacts, and any chats and workspaces you choose to run remotely live only on a bot server you host, if you set one up.
Yes. Every image on this page is the real Maestrly app. Conversations, accounts, bots and their data are demo content prepared for the capture; no real account was used.
Free · Open source · Local-first
Bring your accounts. Choose your models. Add a bot when the work should keep going.