AI Providers and Routing
25+ providers behind one interface, routed automatically.
ZYRAXON is provider-agnostic by design. OpenAI, Anthropic, Google, Groq, OpenRouter, Ollama, Mistral, xAI, Azure, Bedrock and the open-weight ecosystem are all plug-ins behind a single interface. The engine never hardcodes a vendor -- it routes.
Why routing matters
- ▸Price goes up -- the engine switches to a cheaper working model.
- ▸Speed matters -- it switches to the fastest one available.
- ▸A service disappears -- it keeps running on the others.
- ▸You prefer local -- it routes to Ollama or open-weight weights with no subscription.
ℹ
Nobody can switch you off
Because there is no single door to stand behind, a vendor outage or a price change is a routing decision, not an outage. The provider layer lives in
packages/llm and the routing logic in packages/zyraxon/src/provider/.Provider families
| Family | Examples |
|---|---|
| Commercial frontier | OpenAI, Anthropic, Google, xAI, Mistral |
| Cloud-hosted | Azure OpenAI, AWS Bedrock, Google Vertex |
| Aggregators | OpenRouter, Groq |
| Local / open-weight | Ollama, local weights |
| Compatible endpoints | Any OpenAI-compatible base URL |
Bring your own key
Keys are configured once and stored locally. Before anything is sent to an external model, every API key, token and password is filtered out, so a secret in your workspace cannot leak into a prompt. Every tool call asks for your permission first.
Pinned local models
A small set of models can be pinned and run locally -- text, image, video and audio generation -- so a machine with the weights installed keeps working with no network at all.