Workflow Patterns

Core Concepts

• All patterns extend AugmentedLLM

• Unified interfaces ensure composability

• Patterns orchestrate Agents and leverage tools

Available Patterns

AugmentedLLM (Foundation)

• Base for all other workflows

• Standard methods + conversation memory

• Support various LLM providers through implementation classes (OpenAI, Anthropic, etc.)

Parallel Workflow

• Distribute tasks across sub-agents

• Consolidate outputs via fan-in agent

Use Cases: Document analysis, multi-perspective evaluation

Router Workflow

• Route requests to appropriate handlers

• LLMRouter and EmbeddingRouter implementations

Use Cases: Multi-agent systems, intent-based routing

Intent Classifier Workflow

• Specialized classification of user queries

Use Cases: Smart assistants, chatbot understanding

Evaluator-Optimizer Workflow

• Generate → Evaluate → Refine feedback loop

Use Cases: Quality-assured generation, technical content

Orchestrator-Workers Workflow

• High-level planning and task distribution

Use Cases: Complex research, strategic workflows

Swarm Workflow

• Multi-agent collaboration and dynamic switching

Use Cases: Virtual assistants, decision trees