The OpenAI-compatible inference API for the Autonomous AI ecosystem. Point any OpenAI SDK at one base URL, drop in your key, and ship — streaming, tool calls, wallet-native credits, no lock-in.
It's OpenAI-compatible. Change the base URL and key — the rest of your code stays the same. Create a key →
Every model is served through a resilient multi-provider fleet with automatic failover — you get uptime, we handle routing. Common OpenAI ids (like gpt-4o) alias to the closest Jarvis model, so existing code just works.
| Model | Best for | Context |
|---|---|---|
| jarvis-70b | Flagship general model (Llama 3.3 70B class). The default. | 32K |
| jarvis-8b | Fast & cheap for high-throughput / simple tasks. | 128K |
| jarvis-120b | Largest reasoning model for deep tasks. | 32K |
Create a chat completion. Set stream:true for token-by-token SSE. Supports temperature, top_p, max_tokens, stop, tools / tool_choice, and response_format.
List the available models in the standard OpenAI list shape. Requires your API key.
Every request is authenticated with a bearer key created in your console. Keys are shown once and stored hashed — treat them like passwords, keep them server-side.
# header on every request Authorization: Bearer jc_live_...
With stream:true the response is text/event-stream: standard OpenAI chat.completion.chunk frames, terminated by data: [DONE]. Every OpenAI streaming client parses it out of the box.
data: {"choices":[{"delta":{"content":"Hel"}}]}
data: {"choices":[{"delta":{"content":"lo"}}]}
data: [DONE]Requests are metered by token against a prepaid credit balance. Start free with monthly credits, then top up by card or crypto. A request that returns no work is never charged.
Business & Enterprise plans available in the console. Free monthly credits during beta.
Errors use the standard envelope, so your existing error handling works.
{"error":{"message":"...",
"type":"invalid_request_error",
"code":"invalid_api_key"}}