An AI that lives on your hardware.

No cloud. No telemetry. Just a capable assistant with 33 tools, a web UI, a CLI, and a desktop app — entirely under your control.

Python 3.12+·MIT License·GPU optional

Star on GitHub Get Started
Aria web interface

What it does

A tool-driven assistant that picks the right capability for each request.

Tool Registry

+30 Tools in 7 categories. Core tools always loaded; domain tools activated on demand.

Three Interfaces

Web UI via Chainlit, a full CLI for automation, or a native PySide6 desktop app with tray.

Local LLM

GPU inference with vLLM. GPTQ/AWQ quantization for 8 GB+ VRAM. No internet required.

Browser Automation

Headless Lightpanda with CDP support. Navigate, click, and extract — programmatically.

Private by Default

Data stays on your machine. No cloud calls for inference. Complete local control.

Worker Agents

Heavy tasks run in the background. Research, code generation, analysis — non-blocking.

Tool inventory

Core and file tools are always loaded. Everything else activates on demand.

Category Tools Load
Core
reasoningplanknowledgescratchpadweb_searchdownloadweathershell
 Always
Files
read_filewrite_fileedit_filefile_infolist_filessearch_filescopydeleterename
 Always
Browser
open_urlbrowser_click
 On-demand
Development
python
 On-demand
Finance
stock_pricecompany_infoticker_news
On-demand
Entertainment
imdb_searchmovie_detailsperson_detailsfilmographyepisodesreviewstriviayoutube
 On-demand
System
http_requestprocess
 On-demand

Get running

Pick your install method. You'll be chatting in under a minute.

# Clone and run from source
git clone git@github.com:malvavisc0/aria-ai.git
cd aria-ai
uv sync
aria server run
# → http://localhost:9876
# Install from PyPI
pip install aria-ai
aria server run
# → http://localhost:9876
# Run with Docker (GPU required)
docker run -p 9876:9876 \
  -v ./data:/app/data \
  ghcr.io/malvavisc0/aria-ai-cuda:latest
# → http://localhost:9876
# Install with GUI support
pip install aria-ai[gui]
aria-gui
# Or grab the standalone binary from releases

How requests flow

Aria evaluates each message, picks the right tools, and delegates heavy work to background agents.

User
Web / CLI / GUI
→
Aria
Primary Agent
→
Tools
+30 tools, 7 categories
→
Response
Contextual answer
Heavy Tasks
5+ tool calls
→
Worker Agent
Background
→
Result
Async delivery