استخدم مستشار Claude Code للقرارات الصعبة بدون موديل أغلىUse the Claude Code advisor for hard decisions without a pricier model

Key terms

Advisor

A second model, usually stronger, that reads the whole conversation and gives Claude guidance at key moments. It runs server side on Anthropic's infrastructure.

Main model

The model that does the work in the session: reads files, writes code, runs commands.

Pairing

A main model plus an advisor. The advisor must rank at or above the main model.

Prompt cache

Stored conversation prefix that cuts the cost of resending it. Turning the advisor on or off mid session does not clear it.

The steps

1. 1

   Step 1

   Know when the advisor pays off

   Use it on long tasks where most turns are routine but plan quality decides the result. Think large refactors, an error that keeps coming back, or work you want checked before Claude says it is done.

   It adds little on short tasks with nothing to plan. If every turn needs the strongest model, switch the main model instead.

   The feature is experimental and needs the Anthropic API. It is not available on Bedrock, Vertex or Foundry.

   Tip: If DISABLE_TELEMETRY is set, the advisor stays off, because it depends on a feature flag fetched from Anthropic.