Inference API

CoreAI API

Run open models hosted in Uzbekistan. Send JSON requests and receive JSON responses or streamed events.

Create an API key Make a request

Overview

The API uses Bearer authentication and the Chat Completions request and response format.

Base URL

https://inference-api.coreai.uz/v1

Authentication

Bearer API key

Wire formats

JSON · SSE

Quickstart

Create a key, set it as an environment variable, and send your first request.

Create a key in the API console. The plaintext is shown once.
Store it in an environment variable: export COREAI_API_KEY="cai_..."
Send an authenticated POST request to the chat completions endpoint.

curl https://inference-api.coreai.uz/v1/chat/completions \
  -H "Authorization: Bearer $COREAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma4-31b-it",
    "messages": [
      {"role": "user", "content": "Salom!"}
    ]
  }'

Authentication

Include your CoreAI key as a Bearer token on every /v1 request. Keys begin with cai_ and are managed in the API console.

Authorization: Bearer $COREAI_API_KEY
Content-Type: application/json

Never expose an API key in browser code, mobile binaries, logs, or source control. Call CoreAI from your server and load the key from a secret or environment variable.

Models

List the model identifiers currently available to your account.

GET/v1/modelsList available models

curl https://inference-api.coreai.uz/v1/models \
  -H "Authorization: Bearer $COREAI_API_KEY"

Chat completions

Send a conversation as ordered role/content messages. Each request is stateless: include the prior messages your application wants the model to consider.

POST/v1/chat/completionsGenerate a model response

Parameter	Type	Default	Description
model	string	required	Model ID returned by GET /v1/models.
messages	array	required	Ordered system, user, assistant, and tool messages.
stream	boolean	false	Stream response chunks over SSE.
temperature	number	model default	Sampling randomness from 0 to 2.
top_p	number	model default	Nucleus sampling probability from 0 to 1.
max_tokens	integer	model default	Maximum number of output tokens.
stop	string \| array	null	Up to four sequences that stop generation.
seed	integer	null	Best-effort deterministic sampling seed.
reasoning	object	false	Set enabled to true or false. Set exclude to true to omit reasoning text from the response.
tools	array	null	Function definitions the model may call.
tool_choice	string \| object	auto / none	Controls function selection: none, auto, required, or a named function.
parallel_tool_calls	boolean	true	Allow multiple function calls in one response.

Response

{
  "id": "chatcmpl_...",
  "object": "chat.completion",
  "model": "gemma4-31b-it",
  "choices": [{
    "index": 0,
    "message": {
      "role": "assistant",
      "content": "Salom!"
    },
    "finish_reason": "stop"
  }],
  "usage": {
    "prompt_tokens": 9,
    "completion_tokens": 4,
    "total_tokens": 13
  }
}

Streaming

Set stream to true to receive tokens as they are generated rather than waiting for the complete response.

curl -N https://inference-api.coreai.uz/v1/chat/completions \
  -H "Authorization: Bearer $COREAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "gemma4-31b-it",
    "messages": [{"role": "user", "content": "Salom!"}],
    "stream": true,
    "stream_options": {"include_usage": true}
  }'

The response uses text/event-stream. Each frame contains a chat completion chunk. The stream ends with data: [DONE]. With include_usage enabled, the terminal usage chunk contains token totals.

Reasoning

Gemma 4 31B supports thinking on/off. Thinking is off by default. Set reasoning.enabled to true to enable it for a request. Effort levels are unsupported.

curl https://inference-api.coreai.uz/v1/chat/completions \
  -H "Authorization: Bearer $COREAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "google/gemma-4-31b-it",
    "messages": [{"role": "user", "content": "Compare two deployment plans."}],
    "reasoning": {"enabled": true}
  }'

JSON responses return reasoning in choices[0].message.reasoning. Streams return it in choices[0].delta.reasoning, separately from content. The completion token count includes reasoning tokens.

Tool calling

Tool calling is currently unavailable. The examples below show the function-calling request format.

Define functions

curl https://inference-api.coreai.uz/v1/chat/completions \
  -H "Authorization: Bearer $COREAI_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "qwen3.8-27b",
    "messages": [{"role": "user", "content": "What is the weather in Tashkent?"}],
    "tools": [{
      "type": "function",
      "function": {
        "name": "get_weather",
        "description": "Get the current weather for a city",
        "parameters": {
          "type": "object",
          "properties": {"city": {"type": "string"}},
          "required": ["city"]
        }
      }
    }],
    "tool_choice": "auto"
  }'

Function call

{
  "choices": [{
    "message": {
      "role": "assistant",
      "content": null,
      "tool_calls": [{
        "id": "call_abc123",
        "type": "function",
        "function": {
          "name": "get_weather",
          "arguments": "{\"city\":\"Tashkent\"}"
        }
      }]
    },
    "finish_reason": "tool_calls"
  }]
}

Send the function result

{
  "model": "qwen3.8-27b",
  "messages": [
    {"role": "user", "content": "What is the weather in Tashkent?"},
    {
      "role": "assistant",
      "content": null,
      "tool_calls": [{
        "id": "call_abc123",
        "type": "function",
        "function": {
          "name": "get_weather",
          "arguments": "{\"city\":\"Tashkent\"}"
        }
      }]
    },
    {"role": "tool", "tool_call_id": "call_abc123", "content": "{\"temperature_c\":18}"}
  ]
}

Run the function in your application, then send its result as a tool message with the matching tool_call_id.

In a stream, call fragments arrive in choices[0].delta.tool_calls. Concatenate function.arguments fragments in order.

Data retention

Free-tier API prompts, request parameters, and generated responses are retained for up to 30 days and may be reviewed to evaluate quality, understand usage, and prevent abuse.

Content retention

Up to 30 days

Model training

Not included in this consent

API content does not appear in web-chat history or the usage console. Plaintext API keys and request headers are not stored with content records.

Errors and limits

Errors use a stable JSON envelope with a human-readable message, type, parameter, and machine-readable code.

400Invalid request or unsupported field

401Missing, invalid, expired, or revoked API key

403Current Terms of Service acceptance required

404Unknown or unavailable model

429Shared account request limit reached

503Inference capacity temporarily unavailable

Successful and rate-limited responses include x-ratelimit-limit-requests, x-ratelimit-remaining-requests, and x-ratelimit-reset-requests. Honor Retry-After on 429 and 503 responses.