Configuration

The AgentConfig class allows you to fine-tune the behavior of your Agents, such as retry policies, token limits, and logging.

Purpose

Use AgentConfig to: - Set global or per-agent configuration. - Control retry behavior for LLM calls. - Enable parallel tool execution. - Manage context window sizes.

Usage

Applying Configuration

Pass a config object when initializing an Agent.

from yosrai import Agent, AgentConfig

config = AgentConfig(
    max_loops=5,             # Max tool execution loops
    max_retries=3,           # Retries on LLM failure
    timeout=30,              # Request timeout in seconds
    parallel_tool_calls=True, # Execute independent tools in parallel
    thinking=True,           # Enable thinking mode
    reflection=True          # Enable reflection mode
)

agent = Agent(
    name="RobustAgent",
    instructions="Help the user.",
    config=config,
    verbose=True  # Enable step-by-step output
)

Skill Configuration (For Conductors)

When using a Conductor, you can wrap individual skill agents with SkillConfig to enforce specific execution policies like retries and timeouts for that specific skill.

from yosrai import SkillConfig

robust_skill = SkillConfig(
    skill=my_agent,
    retries=3,
    timeout=5.0
)

See Conductor documentation for more details.

API Reference

yosrai.engine.core.config.AgentConfig

Bases: BaseModel

Configuration for the Agent.

Attributes:

Name	Type	Description
model	str	Model identifier (e.g. 'openai/gpt-4o', 'ollama/llama3').
max_turns	int	Maximum number of ReAct loop turns (default: 10).
temperature	float	Sampling temperature (default: 0.0).
api_key	Optional[str]	API Key for the provider.
extra_headers	Dict[str, str]	Extra HTTP headers for requests.