# Sovereign AI

AI has quickly become something you depend on for your business.
The goal with Sovereign AI is that you're in control of your AI stack.

This open source project gives you all the components you need:

- Routing: you can easily swap models as needed for LLM, TTS, STT and more
- Acceleration: a global edge accelerates skills, voice, telephony etc
- Persistence: storage and memory
- Client & Server side SDKs: to quickly build your agents
- Voice: modern voice AI & telephony

This means that you can easily start with OpenAI or Anthropic.
But if at some point you decide to switch to open models or your own models you can do so instantly.

- **Your history and memory**: You control how message history and memory is retained.
- **Any model**: Switching from frontier, to open to fine-tuned models should be easy.
- **Your intellectual property**: How your intellectual property is shared or not shared with labs is configurable.
- **Predictable costs**: Stats on usage to keep costs aligned with your policies

### Why Sovereign AI

There are a few different reasons why companies often prefer sovereign AI

- Control
- Intellectual property, and concerns around sharing with frontier labs
- Costs; lock-in on expensive frontier models
- Quality; your own fine-tuned models can outperform the frontier on specific tasks
- Ease of use and development pace: this stack is the easiest and fastest way to deploy agents

## Architecture for Fast Sovereign AI

Building fast sovereign AI is a fun engineering challenge.
Come say hi on our Discord or Reddit. We always love to hear what people are building and when possible help out.

To build fast sovereign AI we made a few different design decisions:

1. **Agents are defined in a folder and named by their use case**

2. **Fast**

   Regardless if you're building text, video or voice AI you want it to be fast.

   - Go: Its faster, scales to high concurrency and doesn't have weird asyncio issues like Python does
   - Edge network and optimal routing based on STT, TTS locations
   - Custom fine-tuned voice models
   - Fast audio/video join flow with WARP
   - Async voice loops with no turn taking

3. **Frontier & Open**

   It's easy to switch between frontier models, open models on providers like baseten or even your own finetuned models running on your own GPUs.

4. **Use Cases**

   Agents and routers are named by their use case. IE an agent will be called "docs_support", or "healthcare_voice".
   This makes it easy to swap models.

5. **Voice AI**

   The fastest voice AI, powered by modern/ no turn taking voice AI.

6. **Software Factory**

   We run a factory: a set of agents continuously works on this repository together with our team.
   This means the model support and routers are always up to date.

## Community

Join the community and let us know what you're building

- [Discord](https://discord.gg/RkhX9PxMS6): Chat with the team and other builders.
- [Reddit](https://www.reddit.com/r/PLACEHOLDER/): Share what you built and ask questions.
- [GitHub](https://github.com/GetStream/Vision-Agents): Issues, pull requests and the source.

---

For the most recent version of this documentation, visit [https://getstream.io/agents/docs/go-golang/sovereign-ai/](https://getstream.io/agents/docs/go-golang/sovereign-ai/).