Pause and Resume CrewAI Flows

[kuldeepsinghkarki]

Can we pause and resume Crewai flows later on. I tried using flow level persistence and then resumed a flow by passing the flow id, however it always starts from the beginning. Whereas i need the flexibility to start from the method/stage where it was paused. any help/direction would be highly appreacited.

[KeepALifeUS]

This is achievable with external state persistence. Here's the pattern:

State-Based Checkpointing

import json

def save_flow_state(flow_id, state):
    with open(f"checkpoints/{flow_id}.json", "w") as f:
        json.dump(state, f)

def load_flow_state(flow_id):
    try:
        with open(f"checkpoints/{flow_id}.json") as f:
            return json.load(f)
    except FileNotFoundError:
        return None

# Your flow state
state = {
    "flow_id": "flow_123",
    "current_stage": "stage_2",
    "completed_stages": ["stage_1"],
    "stage_outputs": {
        "stage_1": {...}
    },
    "paused": False
}

Resume Logic

def run_flow(flow_id):
    state = load_flow_state(flow_id) or initialize_new_state(flow_id)
    
    # Skip completed stages
    stages = ["stage_1", "stage_2", "stage_3"]
    start_from = stages.index(state["current_stage"])
    
    for stage in stages[start_from:]:
        if state.get("paused"):
            save_flow_state(flow_id, state)
            return "Flow paused"
        
        # Run stage with previous outputs as context
        result = run_stage(stage, state["stage_outputs"])
        state["stage_outputs"][stage] = result
        state["completed_stages"].append(stage)
        state["current_stage"] = stages[stages.index(stage) + 1] if stage != stages[-1] else "completed"
        
        # Checkpoint after each stage
        save_flow_state(flow_id, state)

Why This Works

1. True resume — Start from exact checkpoint, not beginning
2. Previous outputs preserved — Stage 2 has access to Stage 1's results
3. Crash-resilient — Checkpoint after each stage
4. Framework-agnostic — Works with any CrewAI setup

More on state-based workflows: https://github.com/KeepALifeUS/autonomous-agents