Advisor

Not yet available. The Advisor server tool is not yet available.

Advisor watches the conversation as it unfolds and, when it detects a sub-question that a specialist model would answer better than the primary model, it routes that question to the specialist, injects the answer back into the context, and lets the primary model continue. Your application makes one API call and receives one response.

Enabling advisor#

{
  "model": "openai/gpt-5.4",
  "messages": [
    {
      "role": "user",
      "content": "Write a blog post about transformer models, and include a worked Python example of scaled dot-product attention."
    }
  ],
  "tools": [
    {
      "type": "opendunes.advisor",
      "options": {
        "specialists": {
          "code": "anthropic/claude-opus-4.8",
          "math": "google/gemini-2.5-pro"
        }
      }
    }
  ]
}

In this example, when the primary model reaches the code portion of its response, Advisor detects the code intent and routes that sub-task to claude-opus-4.8. The result is injected back into the context before gpt-5.4 continues writing the surrounding prose.

Specialist slots#

All slots are optional. Unmatched sub-tasks stay with the primary model.

Configuration#

Billing#

Each specialist invocation is billed at the specialist model's DA rate. The primary model is billed for the full conversation including injected specialist results. All hops are itemized in the per-request usage breakdown.