◉ ARIA

Autonomous Reasoning and Intelligent Agent

Your project-aware coding partner. Reason before action.

ARIA is an open-source CLI agent that works inside your project boundaries like a senior engineer — it plans before acting, validates its work, and remembers context across sessions.

Demo

◉ aria(my-project) › Build a REST API with FastAPI and test it

  ? What endpoints should the API have?
  Answer: GET /health, POST /items, GET /items

─────────────────── Plan ───────────────────
  Goal: Build FastAPI REST API

  1. Scaffold project with venv + git
  2. Write main.py with endpoints
  3. Install dependencies
  4. Run server and test with curl
  5. Update README

Proceed? (yes / no / modify): yes

  1. new_project  'fastapi-api'
  2. write_file   'main.py'
  3. run_command  '.venv/bin/pip install -r requirements.txt'
     │ Successfully installed fastapi uvicorn
  4. run_command  'curl http://localhost:8000/health'
     │ {"status":"ok"}

─────────────────── Report ─────────────────
✅ API built and tested successfully.
Files: main.py, requirements.txt, README.md
Run: uvicorn main:app --reload

Features

• Plan before action — shows you what it will do, waits for approval
• Validates everything — runs code, reads output, fixes errors automatically
• Project memory — remembers stack, decisions, and context across sessions
• Project isolation — auto-creates git, venv, .env, .gitignore, README
• Approval system — asks before dangerous operations
• Beautiful diffs — shows exactly what changed in every file
• Git tools — commit, diff, branch from natural language
• Streaming responses — real-time output, not a black box

Install

pip install aria-agent

Requirements: Python 3.10+, Ollama (for local/cloud models)

Usage

# Run with default model
aria

# Run with specific model
aria --model nemotron-3-super:cloud

# Run with custom workspace
aria --workspace ~/my-projects

Slash Commands

Models

ARIA works with any Ollama-compatible model:

aria --model nemotron-3-super:cloud   # Ollama cloud
aria --model llama3.3                  # Local via Ollama
aria --model qwen2.5-coder:32b         # Local coding model

How it works

Every task follows this workflow:

Clarify → Plan → Approve → Execute → Validate → Remember → Report

ARIA never silently modifies your system. Every action is visible, every dangerous operation requires your approval.

Project Memory

ARIA stores project context in ~/.aria/:

~/.aria/
├── user_memory.json       # Your preferences
└── projects/
    └── my-project/
        ├── meta.json      # Stack, status, path
        ├── memory.json    # Key decisions
        └── progress.md   # Milestone history

Built by

Sumit — independent developer
Built in collaboration with Claude by Anthropic

License

MIT