domains / agent-tooling / crewai-multi-agent-orchestration / crewai

crewai

Skill card

combined

Both copies describe the same CrewAI framework, but they emphasize different, non-overlapping implementation paths. The Orchestra-Research copy is a code-first guide focused on inline Python definitions, CLI project creation, and core object parameters. The sickn33 copy is a configuration-first guide that emphasizes YAML-based agent/task definitions, the @CrewBase decorator pattern, and specific features like the 'planning' flag and 'manager_llm' configuration. Neither copy is a strict reproduction of the other; combining them provides a more complete picture of how to structure a CrewAI project (either via code or YAML) and which advanced features are available.

Repositories

Other copies

Combined skill

---
name: crewai-orchestration
description: Comprehensive guide to building multi-agent systems with CrewAI, covering both code-first and YAML-configured approaches.
---

# CrewAI Multi-Agent Orchestration

## 1. Setup
Install the framework and tools:
```bash
pip install 'crewai[tools]'
```

## 2. Project Structure
You can define agents and tasks either inline in Python or via YAML configuration files. YAML is recommended for larger projects to separate concerns.

### Option A: Code-First (Inline)
Define agents and tasks directly in your Python script:
```python
from crewai import Agent, Task, Crew, Process

researcher = Agent(
    role="Senior Research Analyst",
    goal="Discover cutting-edge developments",
    backstory="Expert analyst...",
    verbose=True
)

research_task = Task(
    description="Research {topic}",
    expected_output="A detailed report",
    agent=researcher
)
```

### Option B: YAML-Configured (Recommended)
Use `config/agents.yaml` and `config/tasks.yaml` with the `@CrewBase` decorator:
```python
from crewai.project import CrewBase, agent, task, crew

@CrewBase
class MyCrew:
    agents_config = 'config/agents.yaml'
    tasks_config = 'config/tasks.yaml'

    @agent
    def researcher(self):
        return Agent(config=self.agents_config['researcher'])

    @crew
    def crew(self):
        return Crew(agents=self.agents, tasks=self.tasks, process=Process.sequential)
```

## 3. Core Components
- **Agents**: Define `role`, `goal`, `backstory`, and optional `tools`.
- **Tasks**: Define `description`, `expected_output`, and `agent`. Use `context` to pass outputs from previous tasks.
- **Crews**: Group agents and tasks. Choose a `process` type.

## 4. Process Types
- **Sequential**: Tasks run in the order defined. Best for linear workflows.
- **Hierarchical**: An auto-generated manager agent delegates tasks. Use `manager_llm` to specify the model for the manager.

## 5. Advanced Features
- **Memory**: Enable `memory=True` on Agents or Crews for short-term/long-term context.
- **Planning**: Set `planning=True` on the Crew to generate a step-by-step execution plan before kickoff.
- **Flows**: For event-driven, complex state management, consider using CrewAI Flows instead of standard Crews.

## 6. Execution
Run the crew with inputs:
```python
result = crew.kickoff(inputs={"topic": "AI Agents"})
print(result.raw)
```