Maestro
AI Agent Orchestration System
“agents in concert”
A powerful terminal-based AI agent that orchestrates multi-agent workflows with built-in tools, MCP server support, and permission-controlled modes. Built with Bun and TypeScript for the CLI experience that developers actually want.
Conducted, Not Cluttered
Three principles shape everything Maestro does — from the terminal interface to how agents share context.
One Podium
A single conductor for every agent.
Maestro is the baton, not another voice in the room. It coordinates sub-agents, tools, and models from one place — so complex work stays in sync and nothing drifts off tempo.
Every Model, One Stage
Bring any LLM. Switch on a dime.
Claude, GPT, Gemini, Grok, and local Redl GGUFs all perform on the same stage. Change models and providers mid-session with /model — no rebuilds, no lock-in.
Built for the Terminal
A TUI that shows its work.
Streamed responses, inline chain-of-thought, live tool calls, and real diffs — all rendered in your terminal. You always see the hand that moves.
Watch It Conduct
A live look at the terminal. Maestro reads your request, thinks out loud, runs tools, and ships the result.
Install
Get started with Maestro in seconds. Choose your preferred platform.
Windows
irm https://www.viancorp.net/maestro/dist/install-maestro.ps1 | iexLinux
curl -fsSL https://www.viancorp.net/maestro/dist/install-maestro.sh | bashSlash Commands
18 commands to control every aspect of your agent workflow. Press Tab for autocomplete.
| Command | Description |
|---|---|
/brainstorm | Frame turn as brainstorming — multiple perspectives before implementation |
/btw | Ask the LLM a one-shot side question without affecting conversation history |
/clear | Clear the active conversation context (history preserved on disk) |
/compact | Manually compact conversation history to free context space |
/cost | Show token usage and cost for this session |
/debug | Frame turn as debugging — find root cause before fixes |
/explain | Frame turn as explanation — walk through concepts step by step |
/fast | Disable thinking for faster turn responses |
/help | List all available commands |
/init | Generate or refresh MAESTRO.md by scanning the project |
/mode | Change permission mode (plan/ask/auto/yolo) |
/model | Change the LLM model and provider |
/name | Show or set the current session name |
/plan | Enter plan mode — research and propose before executing |
/queue | Inspect or manage the steer queue |
/research | Frame turn as research — prefer web sources with citations |
/resume | Resume a previous session by name or index |
/think | Re-enable thinking for this session |
Quick tip
Use /plan to enter plan mode where the agent researches and proposes before acting,
or /fast to disable thinking for faster responses.
All commands support Tab autocomplete in the TUI.
Features
Everything you need to orchestrate AI agents efficiently — from terminal-native workflows to multi-agent coordination.
Take the Podium
One binary for Windows or Linux. MCP integration, sub-agents, permission modes and every supported model, ready on first launch.