# OpenOnion

> OpenOnion is the open protocol for AI agents — decentralized, peer-to-peer, local-first, and model-agnostic. ConnectOnion is the reference SDK (Python, TypeScript, Rust). Agents run locally, own their data, and connect to any LLM or service without platform lock-in.

## What it is

OpenOnion is a decentralized protocol and an open-source framework for building AI agents. Developers use the ConnectOnion SDK to create agents in 2 lines of code:

```python
from connectonion import Agent
agent = Agent("name", tools=[my_function])
agent.input("what should I do?")
```

Agents communicate peer-to-peer, store data locally, and integrate with any LLM provider: OpenAI, Anthropic, Google Gemini, Mistral, Meta Llama, Ollama, or any OpenAI-compatible API.

## Products

- [ConnectOnion SDK (Python)](https://pypi.org/project/connectonion/) — `pip install connectonion`
- [ConnectOnion SDK (TypeScript)](https://www.npmjs.com/package/connectonion) — `npm install connectonion`
- [ConnectOnion Documentation](https://docs.connectonion.com) — full guides, API reference, tutorials
- [OpenOnion Platform](https://o.openonion.ai) — managed API keys and billing
- [GitHub](https://github.com/openonion) — all source code, MIT licensed

## Pages

- [Home](https://openonion.ai/) — product overview, code example, supported LLMs
- [Manifesto](https://openonion.ai/manifesto) — philosophy: protocols over platforms
- [Careers](https://openonion.ai/careers) — open positions (Founding Engineer, Developer Intern, Marketing Intern)

## Key facts

- Founded: 2024
- License: MIT
- Languages: Python 3.10+, TypeScript / Node.js 18+, Rust
- Default agent model: `co/gemini-2.5-pro` (routes through managed keys)
- Default `llm_do` model: `co/gemini-2.5-flash`
- Max iterations per agent run: 10
- Event hooks: `after_user_input`, `before_llm`, `after_llm`, `before_tools`, `after_tools`, `after_each_tool`

## Common questions

Q: What is the difference between OpenOnion and LangChain / AutoGen?
A: OpenOnion is a protocol first, SDK second. Agents on OpenOnion talk peer-to-peer across networks without a central broker, and data stays with the user. LangChain and AutoGen are agent orchestration frameworks focused on a single process.

Q: Can I bring my own model?
A: Yes. ConnectOnion works with any OpenAI-compatible API. Pass the model string: `Agent(..., model="gpt-4o")` or `model="claude-sonnet-4-6"` or `model="co/gemini-2.5-pro"` (OpenOnion managed keys).

Q: Is data stored on a central server?
A: No. By default, agents run locally and store logs in `.co/logs/` and eval sessions in `.co/evals/`. Peer-to-peer communication uses signed messages; no central database holds user state.

Q: How do I contribute?
A: Read [docs.connectonion.com](https://docs.connectonion.com), then open a PR or issue on [github.com/openonion](https://github.com/openonion). Join the [Discord](https://discord.gg/4xfD9k8AUF).

## Contact

- Email: hi@openonion.ai
- Hiring: openonionai@gmail.com
- Discord: https://discord.gg/4xfD9k8AUF
- X: https://x.com/openonion
- LinkedIn: https://www.linkedin.com/company/openonion/