Skip to main content

Agent Config

Agent configuration controls LLM behavior in agent mode.

Configuration​

Set in manifest.json:

{
  "agent_config": {
    "model": "anthropic/claude-3-5-sonnet-20241022",
    "max_tokens": 4096,
    "max_iterations": 10,
    "system_prompt": "You are a helpful assistant."
  }
}

Parameters​

ParameterTypeDescription
modelstringLLM model (with provider prefix)
max_tokensintegerMax tokens per LLM call
max_iterationsintegerMax tool-calling iterations
system_promptstringAgent system prompt

Supported Models​

Anthropic​

  • anthropic/claude-3-5-sonnet-20241022
  • anthropic/claude-3-opus-20240229
  • anthropic/claude-3-sonnet-20240229
  • anthropic/claude-3-haiku-20240307

OpenAI​

  • openai/gpt-4-turbo
  • openai/gpt-4
  • openai/gpt-3.5-turbo

System Prompt​

The system prompt is auto-generated from your entrypoint's docstring:

@entrypoint
async def customer_support(input: dict, context: AgentContext):
    """
    You are a customer support agent for Acme Corp.

    Capabilities:
    - Answer product questions
    - Track orders
    - Process refunds

    Guidelines:
    - Always be polite
    - Verify user identity before refunds
    - Escalate complex issues
    """
    pass

This becomes the system_prompt in manifest.

Max Iterations​

Limits the agent loop:

{"max_iterations": 10}

After 10 iterations:

  • Agent returns current state
  • Run marked as complete
  • Prevents infinite loops

Token Budget​

{"max_tokens": 4096}

Limits tokens per LLM API call (not total).

Environment Variables​

LLM API keys set via environment:

# Control plane .env
ANTHROPIC_API_KEY=sk-ant-...
OPENAI_API_KEY=sk-...

Or per-run:

POST /runs
{
  "agent_name": "my-agent",
  "llm_api_key_secret": "CUSTOM_API_KEY_SECRET"
}

Next Steps​