Skip to main content
Use ai-stats:advisor when the main model should be able to ask another model for a review, plan, sanity check, or specialist opinion before it finishes. The main model calls Advisor like any other tool. AI Stats runs the Advisor model server-side and returns the advice as the tool result. The main model then writes the final response.
Advisor is gateway-managed across supported text models. It is not converted into Anthropic’s native Advisor tool unless the client explicitly sends Anthropic’s native tool shape.

​
Quick start

curl https://api.phaseo.app/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "openai/gpt-5-nano",
    "messages": [
      { "role": "user", "content": "Design a rate limiter for a distributed API gateway." }
    ],
    "tools": [
      {
        "type": "ai-stats:advisor",
        "parameters": {
          "model": "anthropic/claude-opus-4-8",
          "max_uses": 1,
          "max_completion_tokens": 1400
        }
      }
    ]
  }'

​
Choosing the Advisor model

You can pin the Advisor model in the tool definition: 
{
  "type": "ai-stats:advisor",
  "parameters": {
    "model": "anthropic/claude-opus-4-8"
  }
}
If parameters.model is omitted, the tool call may provide model. If neither is set, AI Stats falls back to the outer request model.

​
Parameters

{
  "type": "ai-stats:advisor",
  "parameters": {
    "name": "reviewer",
    "model": "anthropic/claude-opus-4-8",
    "instructions": "Review plans for correctness, missing edge cases, and implementation risk.",
    "forward_transcript": true,
    "max_uses": 2,
    "max_completion_tokens": 1400,
    "reasoning": { "effort": "high" },
    "temperature": 0.2
  }
}
ParameterTypeDefaultDescription
namestringnoneOptional Advisor name. Names must be unique after trimming and may contain letters, numbers, spaces, underscores, and dashes.
modelstringouter modelAdvisor model to call. If omitted, the tool call can provide model.
instructionsstringdefault Advisor behaviorExtra instructions for the Advisor model.
forward_transcriptbooleanfalseInclude the current conversation transcript in the Advisor request.
max_usesinteger1Maximum calls for this Advisor during the server-tool loop.
max_completion_tokensinteger1400Max output tokens for the Advisor response.
max_tokensinteger1400Legacy alias for max_completion_tokens.
reasoningobjectprovider defaultReasoning config forwarded to the Advisor call when supported by the selected model/provider.
temperaturenumberprovider defaultSampling temperature for the Advisor call.

​
Tool-call arguments

The model normally calls Advisor with a prompt: 
{
  "prompt": "Review this migration plan and identify the riskiest assumptions."
}
If the Advisor definition does not pin a model, the tool call may also include model: 
{
  "model": "anthropic/claude-opus-4-8",
  "prompt": "Check this security design for missing controls."
}
When forward_transcript is true, AI Stats can execute a transcript-only Advisor call if the model does not supply a prompt.

​
Multiple advisors

Add one ai-stats:advisor entry per advisor. Each named advisor becomes a distinct internal tool, such as ai_stats_advisor_security_reviewer or ai_stats_advisor_architect. 
{
  "tools": [
    {
      "type": "ai-stats:advisor",
      "parameters": {
        "name": "security-reviewer",
        "model": "anthropic/claude-opus-4-8",
        "instructions": "Review for vulnerabilities, abuse cases, and missing mitigations."
      }
    },
    {
      "type": "ai-stats:advisor",
      "parameters": {
        "name": "architect",
        "model": "openai/gpt-5",
        "instructions": "Review system design tradeoffs and operational risks."
      }
    }
  ]
}
At most one Advisor entry may omit name. If you force tool_choice: "ai-stats:advisor" with multiple advisors, AI Stats maps that alias to the first configured Advisor.

​
What the tool returns

Advisor returns JSON as the tool result: 
{
  "status": "ok",
  "name": "reviewer",
  "model": "anthropic/claude-opus-4-8",
  "advice": "Start with a smaller migration slice and define rollback criteria before changing traffic routing."
}
If the Advisor request cannot run, AI Stats returns a tool error such as advisor_invalid_request, advisor_max_uses_exceeded, or advisor_request_failed.

​
Conversation memory

Advisor does not keep hidden cross-request state. If you replay prior messages and tool results in the next request, the main model can see the prior consultation. If forward_transcript is enabled, the Advisor can also see the forwarded conversation transcript for that request.

​
Usage and pricing

Advisor calls increment: 
{
  "usage": {
    "server_tool_use": {
      "advisor_requests": 1
    }
  }
}
The Advisor model’s tokens are included in total usage and can be priced at the selected Advisor model’s rates. Server-tool pricing can also use server_tool_advisor_requests.

​
Current limits

  • Advisor sub-agent tools are not enabled yet.
  • Advisor advice streaming is not enabled yet.
  • Recursive Advisor calls are blocked by the server-tool loop limits.
Last modified on June 11, 2026