How to: Customize the Agent Prompt¶

Where to Edit¶

All agent prompts are centralized in:

app/agents/prompts.py

Default Prompt¶

DEFAULT_SYSTEM_PROMPT = """You are a helpful assistant."""

Best Practices¶

1. Be specific about the agent's role¶

DEFAULT_SYSTEM_PROMPT = """You are a customer support agent for agenticos.

Your responsibilities:
- Answer questions about our products and services
- Help users troubleshoot issues
- Escalate complex problems to human support

Tone: Professional but friendly. Use simple language."""

2. Define output format¶

DEFAULT_SYSTEM_PROMPT = """You are a data analyst assistant.

When presenting data:
- Use tables for comparisons
- Include specific numbers and percentages
- Cite your data sources

When you don't know something, say so clearly."""

3. RAG-aware prompt¶

The RAG prompt is in get_system_prompt_with_rag(). It instructs the agent to: - Search the knowledge base before answering - Cite sources using [1], [2] references - List sources at the end of the response

To customize:

def get_system_prompt_with_rag() -> str:
    return f"""{DEFAULT_SYSTEM_PROMPT}

You have access to a knowledge base. Always search before answering.

Rules:
- ALWAYS cite sources: [1], [2], etc.
- If no results found, say so
- Never make up information
- Prefer recent documents over older ones

Sources format:
[1] filename.pdf, page 3
[2] report.docx, page 1"""

4. Multi-persona agents¶

Create different prompts for different use cases:

SUPPORT_PROMPT = """You are a customer support agent..."""
ANALYST_PROMPT = """You are a data analyst..."""
WRITER_PROMPT = """You are a content writer..."""

def get_prompt(persona: str = "default") -> str:
    prompts = {
        "default": DEFAULT_SYSTEM_PROMPT,
        "support": SUPPORT_PROMPT,
        "analyst": ANALYST_PROMPT,
        "writer": WRITER_PROMPT,
    }
    return prompts.get(persona, DEFAULT_SYSTEM_PROMPT)

Tips¶

Keep prompts concise — shorter prompts = faster, cheaper responses
Test with real user queries, not just ideal cases
Include example outputs in the prompt for consistent formatting
Use the temperature setting (AI_TEMPERATURE in .env) to control creativity