07.2 · API

Compute HTTP

OpenAI-compatible chat completions, plus the trust-mode header.

Private Computer issues the API key at pc.0g.ai. The OpenAI-compatible base URL published for 0G hacker environments is router-api.0g.ai/v1. If the dashboard shows a different base URL, the dashboard wins.

Credit pays this call. Gas does not. A 401 is the key. A payment error is the ledger. Trust mode is optional and defaults to standard.

/chat/completions

POSTOne chat completion. The body matches the OpenAI SDK.

Request

POST https://router-api.0g.ai/v1/chat/completions
Authorization: Bearer $ZG_API_KEY
Content-Type: application/json
X-0G-Provider-Trust-Mode: standard

{
  "model": "0GM-1.0-35B-A3B",
  "messages": [
    { "role": "user", "content": "Say hello from 0G." }
  ]
}

Response

{
  "choices": [
    { "message": { "role": "assistant", "content": "Hello from 0G." } }
  ]
}

standard reaches the widest set of models and providers. verified limits routing to attested providers and can return a proof of that route. private is the TEE path for prompts that should not be visible to the provider.
TeeTLS attests routing. TeeML attests that the model ran inside an enclave. Most models are not TeeML. Do not document a blanket “this ran in a TEE.”
Set the header to verified or private only when the model catalog says that mode exists for the model.
The broker alternative is not this HTTP shape. 0g-compute-cli get-secret gives you an app-sk key and a service URL. Call {service}/v1/proxy with that key.