The toolOrchestra is actaully agentOrchestra.

[sunzj]

The "tools" (answer, search, enhance_reasoning) are not really "tools" in the traditional sense (like API calls), but rather they are specialized "agents" that handle specific tasks.

1. Traditional tool: A simple function/API that takes inputs and returns outputs (e.g., calculator, weather API)

2. Agent: A more complex system that can:

   • Take a high-level task
   • Use multiple steps/operations to complete it
   • Have its own internal reasoning process

Looking at what these "tools" do:

• enhance_reasoning: Has a specialized prompt, generates Python code, executes it, returns results - this is an agent performing code generation + code execution
• answer: Has different prompts for different models, extracts answers using specific formats - this is an agent performing question answering with specialized output formatting
• search: Generates search queries, calls retrieval API, extracts results - this is an agent performing search + retrieval

So these tools are more like "sub-agents" or "expert agents" rather than simple tools. The main model acts as an orchestrator that delegates tasks to these specialized expert agents.