Running an agent
AI models
Run the agent on your own API key from the provider you trust.
Orvyn doesn't resell AI. Each agent runs on your own API key, so you choose the model and pay your provider directly.
Supported providers
| Provider | Models |
|---|---|
| Anthropic | Claude |
| OpenAI | GPT and o-series |
| Gemini | |
| xAI | Grok |
| Mistral | Mistral models |
| DeepSeek | DeepSeek models |
| Groq | Fast open models |
| Cerebras | Fast open models |
| Together AI | Open models |
| Fireworks | Open models |
| OpenRouter | 300+ models behind one key |
| NEAR AI | Private inference: Claude, GPT, Gemini, Grok, DeepSeek, Kimi, Qwen and GLM models run inside hardware enclaves (TEEs), and each response comes with a hardware-signed proof |
When you paste a key, Orvyn checks it with the provider and lists the models it can use. Pick one that supports tool use; the agent relies on tools for everything. For OpenRouter and NEAR AI, Orvyn only lists models that support tool use.
NEAR AI
NEAR AI Cloud runs models inside trusted execution environments (Intel TDX and NVIDIA confidential GPUs), so neither the host nor NEAR AI can read your agent's prompts, and every response is attested by the hardware. Create an API key in the NEAR AI Cloud dashboard, paste it in, and pick a model. Orvyn checks the key without running a paid request.
How your key is kept
- Encrypted at rest (AES-256-GCM), bound to the agent, so it can't be copied to another agent's record.
- Never sent back to a browser. The agent page only shows a hint, like
sk-…a1b2. - Used only by the agent's own server, for its autopilot and holder chat.
Changing model or key
On the agent page, open AI model, choose a provider and model, verify, and save. You can keep the stored key when switching models on the same provider.
Use a key dedicated to the agent with a spending limit at your provider. The chat and service limits keep usage in check, and a provider-side cap is a good backstop.