← back to examples
ai intermediate

Runic Structured LLM Branching

Two-phase Runic orchestrator that makes a real deterministic route decision, hot-swaps the branch DAG, and executes the selected workflow locally.

primary showcase ai l2 ai-tool-use runic branching

Related guides and notebooks

Concept
Reasoning strategies compared
Reasoning patterns for branch-heavy decision paths.
Livebook
Reasoning strategies compared
Run the companion notebook in Livebook
Source
branching demo source
Structured route output and branch execution walkthrough.
Source
branching orchestrator module
Route extraction, workflow swap, and phase 2 execution.
Explanation

What this example teaches

Interactive Demo

Run the agent and inspect state

Source Code

Read the production implementation

What you’ll learn

  • How a phase-1 Runic router can select a phase-2 DAG with real runic.set_workflow calls
  • How route metadata (route, detail_level, confidence) can drive both branch choice and UI evidence
  • How to expose direct, analysis, and safe outcomes without any provider or network dependency

Route outcomes

  • :direct -> quick answer branch
  • :analysis -> plan + synthesis branch
  • :safe -> fallback response branch

Workflow shape

RouteQuestion -> runic.set_workflow(route) -> DirectAnswer | AnalysisPlan -> AnalysisAnswer | SafeResponse

The page runs the actual Runic command surface in-process. The route decision is deterministic fixture data, but the branch selection, workflow swap, node execution, and final outputs are real local workflow transitions.

Demo note

No LLM provider, browser session, or remote network call is required. The source tab shows the real local orchestrator, actions, and LiveView used by this page.