Agents

The framework provides two core concrete agent entry points. Choose by execution pattern first, then by control requirements.

Multi-agent orchestration patterns (PlanExecutePattern, ProposeCriticPattern, RouterDelegatePattern) live in the patterns module and are implemented with the same public workflow step primitives available to users.

Overview

• DirectLLMCall

• MultiStepAgent (mode="direct" | "json" | "code")

Decision table

Use case	Recommended pattern	Why
Plain text generation without tools	DirectLLMCall	Lowest orchestration overhead
Iterative direct (no external tools)	MultiStepAgent(mode="direct")	Internal CONTINUE/STOP controller steps
Iterative structured tool loops	MultiStepAgent(mode="json")	ReAct-style multi-step loop with JSON tool_name/tool_input actions
Iterative code-action loops	MultiStepAgent(mode="code")	ReAct-style loop with code actions

Background references

• ReAct

• Toolformer

• MRKL Systems

These references are for conceptual grounding; behavior is defined by the contracts and implementation in this repository.

Examples

• Agent Examples

• examples/agents/README.md

Pages

• Multi-Step Patterns

• Background Reading