OpenRouter
@OpenRouter
New server tool: Advisor Let smaller models consult a higher-intelligence "advisor" model. Helps them escape doom loops, and helps you migrate to cheaper models! 🧵
23retweets363likes
View on XOpenRouter Launches Advisor Tool for Smarter, Cheaper AI Agent Workflows
OpenRouterOpenRouter introduced its new Advisor server tool, enabling AI models to consult higher-intelligence models during complex tasks. This capability helps prevent models from getting stuck in "doom loops" and allows developers to optimize costs by using expensive reasoning only when necessary.
OpenRouter has launched the Advisor server tool, a new capability that allows an AI model to consult a higher-intelligence "advisor model" mid-generation. When an executing model reaches a decision point, gets stuck, or needs help, it can invoke the
openrouter:advisor tool. The advisor model then provides its guidance, unblocking the primary model to continue its task.- Advisor Model Selection
- Any OpenRouter model
- Advisor Capabilities
- Can run as sub-agent with its own tools
- Configuration Options
- Named advisor profiles with custom instructions
- Recursion Protection
- Advisor tool cannot invoke itself
This "advisor strategy" helps AI agents avoid getting trapped in repetitive or failing trajectories, enhancing their overall reliability. It also enables significant cost optimization, as developers can use cheaper models for most operations and only engage more expensive, capable models for critical reasoning steps. Anthropic observed that its Claude Sonnet model, when advised by Opus, improved performance on SWE-bench Multilingual while reducing cost per task.
The Advisor tool can be configured with any OpenRouter model as the advisor, and it can optionally run as a sub-agent with its own tools, such as agentic web tools, to ground its guidance in external information. Developers can also define named advisor profiles with specific instructions and models, allowing the executing model to select the appropriate expert for a given consultation.
Still wondering? A few quick answers below.
The Advisor server tool allows an AI model to consult a more intelligent "advisor model" during its task execution. When the primary model encounters a difficult decision or gets stuck, it can invoke the Advisor to receive guidance and then continue its work.
By using the Advisor tool, developers can employ smaller, cheaper models for most of an agent's operations. The more expensive, higher-intelligence models are only called upon for critical reasoning or when the primary model needs help, reducing overall inference costs.
Yes, the Advisor model can optionally run as a sub-agent with its own tools, such as web search. This allows the advisor to ground its guidance in external information before providing advice to the primary executing model.
Yes, you can configure a roster of named advisor profiles. Each profile can have its own specific model and instructions, allowing the executing model to select a specialized advisor (e.g., a "reviewer" or an "architect") based on the task at hand.
Every HeadsUpAI update is written based on its original source and reviewed before it's published. Read our editorial standards →
