Help &
Resources.

Yes. The Mycelis proxy endpoint is fully OpenAI-compatible. You only need to set base_url to https://api.mycelis.ai/proxy/v1 and api_key to your PAT. No SDK changes needed.

In the model marketplace you can deploy common open-source models: Llama 3.1 (8B, 70B, 405B), Mistral, Mixtral, Qwen, and more. For Managed Keys, OpenAI (GPT-4o, o1), Anthropic (Claude 3.5), and Google (Gemini 1.5) are available.

Your API requests are processed only on your own GPU instance. Inference runs on dedicated hardware — no data is forwarded to model providers. Prompts and responses are not stored permanently.

With Managed Keys, you pay pay-per-token via Mycelis — no own API key at OpenAI/Anthropic required. With BYOK (Bring Your Own Key), you provide your own API key; traffic is routed and logged via Mycelis, but costs are billed to your own provider account.

Smart Routing is part of the VirtualModel system. You define rules (e.g., 'requests with <500 tokens → small model, otherwise → large model') and Mycelis automatically selects the right deployment. This can reduce API costs by up to 80% without changing your code.

New accounts receive a small starting credit to try the platform. After that, usage is fully pay-as-you-go — no monthly base fee, no minimum commitment.