# Andromity

> A terminal AI coding agent. Autonomous by choice, gated by trust.

Andromity is a terminal workspace for AI-driven software engineering. Rather than a chat sidebar or plugin, Andromity provides a terminal workspace with native session management, diff viewers, a cron scheduler, specialized agent profiles, MCP tool integration, and multi-model flexibility.

- Website: https://andromity.agenticmarket.dev
- Repository: https://github.com/agenticmarket/andromity
- PyPI: https://pypi.org/project/andromity/

## Installation

```bash
# Linux / macOS
curl -fsSL https://raw.githubusercontent.com/agenticmarket/andromity/main/install.sh | bash

# Windows (PowerShell)
irm https://raw.githubusercontent.com/agenticmarket/andromity/main/install.ps1 | iex

# Python pipx (any platform with Python 3.11+)
pipx install andromity
```

## Quick Start

```bash
# Interactive TUI mode in current workspace
andromity

# Open specific directory
andromity /path/to/project

# Headless / scripted run
andromity run "add error handling to auth.py"

# Unattended run (auto-approve all safe operations)
andromity run "refactor this to async" --yes

# Dry run (simulate changes without modifying files)
andromity run "review session.py" --dry-run
```

## Core Architecture & Features

- **Folder Trust Model**: Gated execution. When entering an untrusted workspace, Andromity refuses to write files or execute commands until explicit permission is granted.
- **Permission Modes**:
  - `SAFE`: Approve plans, file modifications, and commands individually.
  - `TRUST`: Plans reviewed; safe file writes and commands proceed directly.
  - `FULL`: Direct autonomous execution.
  - `YOLO`: Fully unattended execution (ideal for headless cron workflows).
- **Built-in Cron Scheduler**: `/cron` command configures scheduled or interval background tasks saved in `.andromity/crons.json`.
- **Agent Profiles**:
  - `builder`: Plans first, then implements changes.
  - `coder`: Immediate implementation without preliminary planning.
  - `reviewer`: Read-only code audit and findings generator.
  - `planner`: Creates high-level architectural proposals without file writes.
- **Model Agnostic**: Powered by LiteLLM. Compatible with Claude (Anthropic), GPT-4o (OpenAI), Gemini (Google), Groq, DeepSeek, local Ollama, and NVIDIA NIM. Switch models mid-session using `Ctrl+L`.
- **Model Context Protocol (MCP)**: Supports `mcp.json` specifications for connecting custom tools and databases.
- **Diff Viewer & Undo**: Visual git-style diffs with `/undo` turn rollback.

## Full Technical Documentation

For complete technical specifications, command flags, and configuration schemas:
- [Full LLM Documentation](https://andromity.agenticmarket.dev/llms-full.txt)