Routing (Rule-based / LLM-based)

What Problem It Solves

When you have multiple task types, a single prompt/pipeline becomes a compromise. Routing chooses the best specialized flow for the input.

When to Use

• Distinct intents (math vs writing vs retrieval vs code).
• Different cost/latency budgets per route.
• You want explicit control over “what happens next”.

Core Flow

flowchart TD
  U["User input"] --> R["Router"]
  R --> A["Route A (workflow/agent)"]
  R --> B["Route B (workflow/agent)"]
  R --> C["Route C (workflow/agent)"]
  A --> O["Output"]
  B --> O
  C --> O

Evolution Path

• Comes from: Prompt Chaining (multiple workflows exist)
• Leads to: Handoff / Multi-agent (routing between agents), Agentic RAG (route to retrieve)

Repo Reference

• Code: src/agent_patterns_lab/patterns/routing.py
• Example: examples/12_routing.py
• Tests: tests/test_routing.py