​
Chat Completions

POST /v1/chat/completions is an OpenAI-wire-compatible proxy. If you have code that talks to the OpenAI Chat Completions API, point its base URL at AIUS, use an aius_… bearer token, and pass an Anthropic Claude model slug — it just works. The provider key is held server-side; you never send one. 
POST https://aius.co/api/v1/chat/completions

​
Request

​
Headers

Authorization: Bearer aius_xxxxxxxx...
Content-Type: application/json

​
Body

The body is forwarded to the upstream model in OpenAI Chat Completions shape. 
{
  "model": "anthropic/claude-sonnet-4",
  "messages": [
    { "role": "system", "content": "You are a concise assistant." },
    { "role": "user", "content": "Say hello in one sentence." }
  ],
  "stream": false,
  "max_tokens": 256,
  "temperature": 0.7
}
FieldTypeRequiredNotes
modelstringYesAnthropic Claude slug, e.g. anthropic/claude-sonnet-4. Call GET /v1/models for the live list.
messagesarrayYesOpenAI message objects (role + content).
streambooleanNotrue streams Server-Sent Events. Default false.
max_tokensintegerNoUpper bound on generated tokens.
temperaturenumberNoSampling temperature.
toolsarrayNoOpenAI-style tool/function definitions for tool calling.
Standard OpenAI Chat Completions fields (top_p, stop, tool_choice, etc.) are forwarded upstream. The proxy does not strip them.

​
Non-streaming response

Returns the standard Chat Completions object: 
{
  "id": "chatcmpl-...",
  "object": "chat.completion",
  "created": 1717123456,
  "model": "anthropic/claude-sonnet-4",
  "choices": [
    {
      "index": 0,
      "message": { "role": "assistant", "content": "Hello there!" },
      "finish_reason": "stop"
    }
  ],
  "usage": { "prompt_tokens": 18, "completion_tokens": 4, "total_tokens": 22 }
}
The proxy also echoes correlation headers on the response so you can tie a call to a run-model trace: x-aius-session-id, x-aius-run-id, x-aius-step-run-id.

​
Streaming response

Set "stream": true. The response is text/event-stream of OpenAI chunk objects, terminated by data: [DONE]: 
data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"role":"assistant"},"finish_reason":null}]}

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":"Hello"},"finish_reason":null}]}

data: {"id":"chatcmpl-...","object":"chat.completion.chunk","choices":[{"index":0,"delta":{"content":" there!"},"finish_reason":null}]}

data: [DONE]

​
Examples

​
Non-streaming

curl -X POST https://aius.co/api/v1/chat/completions \
  -H "Authorization: Bearer aius_xxxxxxxx..." \
  -H "Content-Type: application/json" \
  -d '{
    "model": "anthropic/claude-sonnet-4",
    "messages": [{"role": "user", "content": "Say hello in one sentence."}]
  }'

​
Using the OpenAI SDK

Because the wire format matches, the official OpenAI SDKs work directly: 
from openai import OpenAI

client = OpenAI(
    base_url="https://aius.co/api/v1",
    api_key="aius_xxxxxxxx...",
)
resp = client.chat.completions.create(
    model="anthropic/claude-sonnet-4",
    messages=[{"role": "user", "content": "Hello!"}],
)
print(resp.choices[0].message.content)

​
Streaming

import httpx

with httpx.stream(
    "POST",
    "https://aius.co/api/v1/chat/completions",
    headers={"Authorization": "Bearer aius_xxxxxxxx..."},
    json={
        "model": "anthropic/claude-sonnet-4",
        "messages": [{"role": "user", "content": "Tell me a short story."}],
        "stream": True,
    },
    timeout=None,
) as r:
    for line in r.iter_lines():
        if line.startswith("data: ") and line != "data: [DONE]":
            print(line[len("data: "):])

​
List models

GET /v1/models (bearer required) returns the OpenAI-style model list the gateway will accept: 
curl https://aius.co/api/v1/models \
  -H "Authorization: Bearer aius_xxxxxxxx..."

​
When to use the run loop instead

/v1/chat/completions is a stateless proxy: you own the conversation, the tool loop, and tool execution. If you want the server to drive the agent loop and just hand you tool calls to execute locally, use the run-loop WebSocket instead.