stigmergy-crewai 0.2.0

Stigmergy pressure-field scheduling adapter for CrewAI

stigmergy-crewai

Optimal task ordering for CrewAI crews. Declare tasks with priorities and dependencies, and stigmergy computes the best execution order using pressure signals. No manual ordering required.

CrewAI runs tasks sequentially -- the order you define them is the order they run. When you have complex dependency graphs, getting that order right by hand is tedious and fragile. Stigmergy resolves the optimal order from your dependency graph automatically, and handles priority-based reordering when tasks compete.

5 lines to integrate. Works with your existing CrewAI agents.

Install

pip install stigmergy-crewai

Usage

from crewai import Agent
from stigmergy_crewai import StigmergyManager

# Create CrewAI agents
researcher = Agent(role="researcher", goal="Find information", llm=llm)
writer = Agent(role="writer", goal="Write content", llm=llm)
editor = Agent(role="editor", goal="Edit content", llm=llm)

# Create stigmergy manager with tasks and dependencies
manager = StigmergyManager(wake_threshold=0.4)
manager.add_task("research", agent=researcher, description="Research AI trends",
                 expected_output="Key findings", priority=0.8)
manager.add_task("write", agent=writer, description="Write article",
                 expected_output="Draft article", priority=0.6, deps=["research"])
manager.add_task("edit", agent=editor, description="Edit the draft",
                 expected_output="Final article", priority=0.5, deps=["write"])

# Preview execution order
print(manager.task_order)  # ['research', 'write', 'edit']

# Build and run
crew = manager.build_crew(verbose=True)
result = crew.kickoff()

How It Works

Since CrewAI's Process enum is not extensible, the StigmergyManager:

1. Accepts tasks with priorities and dependency graphs
2. Uses pressure signals to resolve optimal execution order
3. Deposits completion signals that propagate to dependent tasks
4. Builds a CrewAI Crew with tasks in the resolved order
5. Uses CrewAI's context parameter to pass outputs between dependent tasks

Limitations

• Cannot parallelize tasks (CrewAI sequential process limitation)
• Cannot dynamically inject tasks mid-execution
• Task ordering is resolved at build_crew() time, not runtime

For full dynamic scheduling, use the core stigmergy library directly.

License

MIT