Prompts

The system prompt defines an agent’s role and behavior: what it does, which tools it uses and how it answers.

Configuration

The agent system prompt is a plain string declared as the SYSTEM_PROMPT class variable of the agent, which is required. The framework passes it on as AgentConfig.system_prompt, see Agent Configuration:

# prompt.py (conventional location)
ARTICLES_SYSTEM_PROMPT = """
Answer from the articles you find with search_articles.
"""

# agent.py
class ArticlesAgent(BaseToolCallingAgent):
    NAME = "Articles"
    AGENT_KEY = "articles-agent"
    SYSTEM_PROMPT = ARTICLES_SYSTEM_PROMPT
    TOOLS = [LLMToolConfig(tool_class=SearchArticlesTool)]

An agentic execution can use a different prompt by declaring its own SYSTEM_PROMPT on its definition, see Agentic Execution.

Template Variables

The system prompt can contain {variable} placeholders. _get_system_prompt_variables() returns the variables that fill them, and it is called each time the prompt is built.

By default it returns the fields of the agent’s PROMPT_INPUT by name, filled with the values entered in the UI, see Agent Classes. With this PROMPT_INPUT and prompt, {output_format} needs no override:

class PromptInputSchema(RequiredFieldsModel):
    output_format: str = Field(title="Output Format", description="Expected response format")
 
SYSTEM_PROMPT = """
You answer questions about articles. Format every answer as {output_format}.
"""

To rename variables or compute others, override the method in the agent class and call super() to keep the defaults:

SYSTEM_PROMPT = """
You answer questions about articles. Today is {today}. Format every answer as {output_format}.
"""
 
class ArticlesAgent(BaseToolCallingAgent):
    def _get_system_prompt_variables(self) -> dict:
        return {**super()._get_system_prompt_variables(), "today": date.today().isoformat()}

Agents with a static prompt and no PROMPT_INPUT fields need no override.