HierarchicalPlanningMode
The "Decompose & Delegate" engine. Breaks complex objectives into a recursive tree of sub-goals and parallel tasks with automated dependency resolution.
Tree Architecture
Parallel Execution
Stateful Persistence
⚙️ OrchestrationConfig.json
{
"mode": "HierarchicalPlanningMode",
"maxConcurrency": 4,
"maxIterations": 200,
"model_config": {
"smart": "gpt-4-turbo",
"fast": "gpt-4o-mini"
}
}
→
🌳 Live Goal Tree
🟢 G1: Build Web Scraper
✅ G2: Research Target Site
⏳ G3: Implement Parser Waiting: G2
🏃 T1: Setup Project Running
[System] Iteration 4: Decomposing G3...
[System] Submitting T1 to FixedConcurrencyProcessor...
[System] Submitting T1 to FixedConcurrencyProcessor...
The Decompose-Delegate Loop
The mode operates in a continuous loop until the root goal is achieved or a stop is requested.
- Decomposition: Active goals are sent to the LLM to be broken down into
SubGoalsor executableTasks. - Dependency Resolution: The engine calculates the dependency graph. Tasks are only moved to
PENDINGonce all prerequisite IDs areCOMPLETED. - Parallel Execution: A
FixedConcurrencyProcessorpulls pending tasks and executes them across available worker threads. - Reflection: As tasks complete, the engine re-evaluates the status of parent goals.