Skip to content

Prompts in LLMling-Agent

LLMling-Agent provides a flexible system for handling different types of prompts, allowing seamless conversion of various data types into prompt text.

Prompt Types

The following types can be used as prompts:

  • Strings: Plain text prompts
  • Objects implementing __prompt__: Dynamic content providers
  • Types implementing __prompt_type__: Type-level information
  • Pydantic models: Structured data
  • Base prompts: LLMling-Agent prompt types (Static, Dynamic, File)
  • Dictionaries: Key-value representations
  • Lists/Tuples: Sequences of prompts
  • Callables: Functions returning prompts
  • Coroutines: Async functions returning prompts

Prompt Protocols

__prompt__ Protocol

The __prompt__ protocol allows objects to provide LLM-friendly descriptions of themselves. When implementing this method, follow these guidelines:

Best Practices

Be Descriptive Yet Concise 

class DataAnalysis:
    def __prompt__(self) -> str:
        return f"Analysis of {self.dataset}: {self.key_findings[:200]}..."
        # ✓ Includes context and main content
        # ✓ Truncates long content

Format for LLM Understanding 

class APIEndpoint:
    def __prompt__(self) -> str:
        return f"""
        REST Endpoint: {self.path}
        Method: {self.method}
        Purpose: {self.description}
        """.strip()
        # ✓ Clear structure
        # ✓ Key information first
        # ✓ Human-readable format

__prompt_type__ Protocol

Types can implement the __prompt_type__ classmethod to provide type-level information:

class ResponseType:
    @classmethod
    def __prompt_type__(cls) -> str:
        return "A structured response with fields: ..."

Recursive Resolution

Prompts are resolved recursively: 

await agent.run([
    "Basic text",
    data_source,        # __prompt__ called
    ResponseType,       # __prompt_type__ called
    {"key": "value"},   # Formatted as key-value
    my_prompt_func,     # Called to get result
])

All components are combined into a single coherent prompt.

Supported in LLMling-Agent

The prompt system is used throughout LLMling-Agent:

  • Agent.run() accepts any prompt type
  • Agent instances implement __prompt__
  • Context loading supports all prompt types
  • Tasks can use any prompt type
  • Knowledge sources can be any prompt type

For more examples and detailed API documentation, see the API reference.