Martini 🍸 Now available via Homebrew

Martini is not just a terminal wrapper—it is a professional-grade Zsh orchestrator and autonomous agent built in Swift. By leveraging Apple Intelligence, Martini moves beyond static aliases to research, reason, and act within your macOS environment to fulfill complex intents.

View on GitHub

Join Martini Discord

The Vision: Your Computer, Automated

Martini is the new way to have a dedicated agent on your computer. We are building toward a world where your intent translates directly into system-level action:

Agentic Automation: You no longer write scripts; you describe a goal. Martini will synthesize the logic, verify the requirements, and build the automation for you.
Native Mac Integration: Martini bridges the gap between the shell and macOS. You can command Martini to create a specific automation, and it will generate the native macOS automations or Shortcuts for you.
The CLI App Store: We are evolving into a platform for agent-ready CLI applications, allowing users to deploy specialized agentic behaviors for any professional workflow.

Agentic Features

Autonomous Research: When faced with an unknown tool or complex syntax, Martini proactively uses the ManualLookup tool to ingest official documentation and self-correct.
Persistent Contextual Awareness: Unlike standard shells, Martini maintains a memory of your intent and history, allowing it to execute multi-step plans across different sessions.
Stateful Navigation: The agent understands where it is and where it needs to go, using FileDiscovery and ChangeDirectory tools to explore your filesystem intelligently.
Safety-First Reasoning: Before any high-risk action (like rm or sudo), Martini’s safety engine evaluates the command, flags it, and presents a formal proposal for your approval.

How the Agent Operates

Martini follows a cognitive loop to ensure accuracy and sovereignty:

Deconstruct: It analyzes your natural language intent to identify the necessary system tools.
Synthesize: It researches official “man” pages to ensure the proposed logic is perfect.
Propose: It presents a detailed breakdown of flags and side effects, acting as a transparent collaborator.
Execute & Learn: Upon confirmation, it executes the task via /bin/zsh and updates its internal history to inform future actions.

Technical Foundation

Intelligence: Powered by Apple Intelligence (SystemLanguageModel).
OS Requirements: Optimized for macOS 26.2+.
Language: Built with Swift 5.0 for native performance and safety.

Installation & Deployment

brew install martini

martini

License

Proprietary / Non-Commercial Private Use.