Skip to main content

Local-first AI agent with memory, knowledge graph, tools, web dashboard, and Telegram (OpenYF AI).

Project description

ARIA by OpenYF AI

Building intelligent systems. One phase at a time.

ARIA is an open source AI agent that lives on your machine, remembers everything you tell it, builds a knowledge graph of what it learns, improves itself over time, and works 24/7 — even while you sleep.

Not a chatbot. A cognitive agent with persistent memory, a world model, tool use, and a self-improvement loop. Built openly by YF as the first step toward general intelligence.

PyPI License Python


What makes ARIA different

Most AI tools forget you the moment the conversation ends. ARIA doesn't.

  • Persistent memory — remembers everything across sessions, forever, stored locally on your machine
  • Knowledge graph — builds a causal map of what it learns, not just a list of facts
  • Self-improvement loop — critiques its own responses and retries when unsatisfied
  • Tool use — web search, file access, code execution, browser
  • Multi-agent debate — two instances argue opposing views to find truth
  • Telegram integration — lives in your phone, works while your laptop is closed
  • Your data, your machine — nothing leaves your device

Install

pip install openyfai-aria
aria setup

Open http://127.0.0.1:8000 and start talking.

Windows users: If aria is not recognized after install, close and reopen your terminal. If blocked by Device Guard, run python -m scripts.cli web instead.


The roadmap

ARIA is being built in 7 phases, in public, with full documentation of every decision.

Phase What it builds Status
01 Cognitive core + persistent memory ✅ Complete
02 World model — causal knowledge graph ✅ Complete
03 Self-improvement loop ✅ Complete
04 Multi-agent debate ✅ Complete
05 Tool use + embodiment ✅ Complete
06 Transfer + generalization ✅ Complete
07 Recursive self-improvement 🔄 In progress

Full roadmap: docs/planning/ROADMAP.md


More ways to run

Path When to use
Web UI (aria web) Default: finish setup in the browser.
Terminal agent (aria after setup) Keyboard-first use.
CLI setup only (aria setup) You prefer not to use the web wizard.
Docker Deploy to a server or share on a LAN. See docs/quickstart.md.

Provider support

Gemini, OpenAI, Anthropic, OpenRouter, DeepSeek, Mistral, Groq, Ollama, and local OpenAI-compatible endpoints. Choose your provider during aria setup or change it anytime in settings.


Skills

Add a Markdown file under skills/, restart ARIA, and the agent can use that workflow without touching Python. Start with skills/repo_researcher.md or skills/README.md.


Safe defaults

ARIA ships locked down. Nothing runs without your approval.

  • Tool approvals are on
  • Terminal execution is off
  • Direct code writes are off
  • Background actions are off
  • Remote web access requires ARIA_WEB_API_KEY
  • Telegram is deny-by-default until paired from Operator

Read SECURITY.md before exposing ARIA beyond localhost.


Requirements

  • Python >=3.11
  • Node 20+ — only if you rebuild aria-ui/ from source (not needed for pip install)
  • Docker — optional

Contributing

ARIA is built in public. Every decision is documented, every phase is open. Contributors welcome — see CONTRIBUTING.md and the seeded issues in docs/planning/launch-issue-backlog.md for good first tasks.

Follow the journey on X: @openyfai


Repository guide

Path What lives here
aria/core/cognitive_loop.py Core cognitive pipeline
aria/llm/ LLM providers
aria/memory/ Memory store + retrieval
aria/world/ Knowledge graph + hypotheses
scripts/web_server.py Web backend (FastAPI)
scripts/cli.py CLI entrypoint
scripts/telegram_bot.py Telegram integration
aria-ui/src/ Web UI (Next.js)
skills/ Markdown skill definitions

Development

git clone https://github.com/openyfai/aria.git && cd aria
pip install -e ".[dev]"
python -m pytest tests
ruff check .

To rebuild the web UI from source:

cd aria-ui && npm install && npm run build && cd ..

Project docs


License

Apache-2.0. See LICENSE.


OpenYF AI — building toward general intelligence, in public.

Project details


Download files

Download the file for your platform. If you're not sure which to choose, learn more about installing packages.

Source Distribution

openyfai_aria-1.0.5.tar.gz (617.4 kB view details)

Uploaded Source

Built Distribution

If you're not sure about the file name format, learn more about wheel file names.

openyfai_aria-1.0.5-py3-none-any.whl (634.6 kB view details)

Uploaded Python 3

File details

Details for the file openyfai_aria-1.0.5.tar.gz.

File metadata

  • Download URL: openyfai_aria-1.0.5.tar.gz
  • Upload date:
  • Size: 617.4 kB
  • Tags: Source
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for openyfai_aria-1.0.5.tar.gz
Algorithm Hash digest
SHA256 7f2de3ae8db7b64243a529241ba8eeb94e49190043bb63361b08350583957c86
MD5 a7de57c6e4d0f65d6893f544f2013ddd
BLAKE2b-256 d16ae972ff1b38854def93a37547cf7279aa6d1a1195751fbf539ec76797dd7b

See more details on using hashes here.

File details

Details for the file openyfai_aria-1.0.5-py3-none-any.whl.

File metadata

  • Download URL: openyfai_aria-1.0.5-py3-none-any.whl
  • Upload date:
  • Size: 634.6 kB
  • Tags: Python 3
  • Uploaded using Trusted Publishing? No
  • Uploaded via: twine/6.1.0 CPython/3.13.12

File hashes

Hashes for openyfai_aria-1.0.5-py3-none-any.whl
Algorithm Hash digest
SHA256 ad6febec61f382cc69626def64bce786f52bc4edcdd1c31959710b5071ca5f64
MD5 208938f78dd8595c654f82af9b1c2353
BLAKE2b-256 96409e2c7b2f4abd5f654b634299d46a57d9a0d8f53b66b51884d06f4e56c071

See more details on using hashes here.

Supported by

AWS Cloud computing and Security Sponsor Datadog Monitoring Depot Continuous Integration Fastly CDN Google Download Analytics Pingdom Monitoring Sentry Error logging StatusPage Status page