AIVAX Console

One endpoint. Shared AI behavior.

Applications keep one OpenAI-compatible request. Behind it, the model, knowledge, tools and policy can change as one shared system.

See the request anatomy

Two operating modes

Choose where the complexity should live.

Direct model

Keep a small call small.

Use an integrated model tag when the application owns the whole request and nothing needs to be shared yet.

Find an AI model
AI Gateway

Let shared behavior live behind the call.

Change instructions, retrieval, tools, moderation or provider settings without redeploying every application.

Read the gateway guide

Request anatomy

Watch one call branch into a system.

The model identifier can point directly to an integrated model or to a gateway that carries the reusable behavior around it.

Behind the endpoint Gateway composition
  • Instructions
  • RAG
  • Tools
  • Output
  • Policy
  • Context
01 / Client request POST /v1/chat/completions 
curl https://inference.aivax.net/v1/chat/completions \
  -H "Authorization: Bearer $AIVAX_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<gateway-slug>",
    "messages": [
      {
        "role": "user",
        "content": "Checkout timed out after the database pool reached capacity. Return severity, summary and next_action."
      }
    ],
    "response_format": { "type": "json_object" }
  }'

02 / Structured result The incident, already shaped.
application/json
Severity High
Summary
Checkout timeouts followed database pool saturation.
Next action
Inspect slow queries and database pool wait time.
Prompt
318 tokens
Completion
64 tokens
Total
382 tokens

Response fields and values are illustrative.

Trace kept with the result

Model, usage, errors and attached resources stay together.

Every answer leaves an operational trail.

When conversation logging is enabled, follow the origin, request ID, actual model, usage, linked resources and error state in one record.

Illustrative conversation record requestId: req_...

One request, connected end to end.

  1. 01
    Origin

    Direct API

    OpenAI-compatible request
  2. 02
    Resolved route

    <gateway-slug>

    model: <resolved-model>
  3. 03
    Outcome

    Completed

    318 prompt · 64 completion · no error
Context kept beside the run

Gateway · API key · attached collection

Let an MCP client call the gateway.

Publish one configured gateway as a callable tool without exposing the prompt, retrieval or provider setup behind it.

https://inference.aivax.net/v1/mcp/inference
Authorization
Bearer API key
X-Mcp-Model-Name
Published tool name
Read AI Gateway docs

Change the system behind the endpoint, not every application.

Build in Console