FAQ¶

General¶

Pauhu is a framework for building AI systems. Define what you want the AI to do, and Pauhu handles the prompting and orchestration.

Pauhu focuses on declaring behavior rather than chaining prompts. You describe inputs and outputs; Pauhu figures out how to get there.

Any LLM with an API: OpenAI, Anthropic, Google, local models via Ollama, and more.

Use Predict for simple, fast tasks. Use ChainOfThought when the task benefits from step-by-step reasoning (math, analysis, complex decisions).

Optimizers take your module and training examples, then find better prompts or few-shot examples that improve performance on your metric.

Yes, but you usually don't need to. Pauhu generates effective prompts from your signatures. If needed, you can customize instructions.

Use modules for single tasks (translate, classify, extract). Use agents for workflows that require decisions, tool use, and multiple steps.

Decorate functions with @function_tool:

@function_tool
def my_tool(query: str) -> str:
    """Describe what the tool does."""
    return do_something(query)

Yes. Use .as_tool() to make an agent callable by another agent, or use handoffs for direct transfers.

Yes. Pauhu is designed for production use with built-in error handling, retries, and observability.

Yes. Pauhu includes features for transparency, logging, and human oversight that support EU AI Act compliance.

Pauhu handles rate limits automatically with exponential backoff. You can also configure custom rate limiting.