Quick-start guide for building multi-agent crews in Forjinn with CrewAI. Assign roles, split tasks, and coordinate AI agent teams.

CrewAI: How to Use and Available Options

CrewAI builder showing the agent configuration interface with role assignment, tool selection, and crew coordination options

CrewAI enables you to build powerful, coordinated teams of AI agents who tackle goals together — assigning roles, splitting tasks, and blending strengths inside easy-to-design workflows on Forjinn.

Quick Start: Creating a CrewAI Workflow

Go to your workspace and select CrewAI from the dashboard or add a Crew node in your flow editor
Click Add Agent to bring in supervisors, workers, or custom roles
Enter instructions and settings for each agent (e.g., assign Research, Write, Critique roles)
Link agents together to define "who delegates what" and completion order (sequential, parallel, or custom)
Assign tools or shared resources by clicking the options menu next to each agent
Save and run your workflow — monitor in real time as agents communicate, solve, and hand off tasks

Available Options for Users

Add as many agents/roles as you need (supervisor, worker, analyst, researcher, writer, editor, etc.)
Choose connection type: Sequential (A → B → C), Parallel (all at once), or Hybrid
Assign shared tools to all or any agents (search, API access, file readers, code interpreters, etc.)
Set completion triggers and criteria (when is a task finished? who collects results?)
Adjust agent instructions, triggers, and workflow direction at any time
Enable shared memory for context persistence across the crew

What is CrewAI?

CrewAI provides:

A system for defining crews (groups) of specialized agents that interact over defined protocols
Agent role assignment: supervisor, worker, planner, analyst, critic, researcher, writer, editor, and more
Workflow decomposition — breaking big goals into agent-assigned subproblems, then reassembling final solutions
Communication protocols — messages, signals, and status updates enable dynamic chaining, error recovery, and task escalation
Shared resources — tools, memory, and context available to the entire crew or specific subsets

CrewAI Configuration in Forjinn

Crew Definition: Build a Crew node specifying which agents/roles are part of the crew and their responsibilities
Inter-Agent Communication: Built in with the platform message bus — agents send, receive tasks, results, and signals
Supervisor/Worker: Use dedicated SupervisorAgent and WorkerAgent nodes for top-down task delegation
Parallel/Sequential Flows: Connect multiple worker agents for batch processing or chain them for stepwise goal solving
Shared Tools/Resources: Assign tools to all agents or selectively to specific roles
Memory Settings: Toggle shared memory for context retention across the crew

Example: Content Generation Crew

SupervisorAgent: Receives the main task, breaks it into research, writing, and review sub-tasks
WorkerAgent 1 (Researcher): Uses search tools for data gathering and source collection
WorkerAgent 2 (Writer): Turns research into structured drafts
WorkerAgent 3 (Editor): Reviews output, flags errors, requests rewrites
Result: Coordinated problem solving with human-like teamwork, fully traceable via the platform UI

Using CrewAI Patterns

Set up roles with clear instructions using agent settings in the sidebar
Connect agents to the communication bus: Default out of the box, or use explicit message nodes for custom routing
Configure completion triggers: When all workers finish, the supervisor assembles results and passes output forward
Monitor Execution: Use Agent Executions and Trace panels — each sub-agent's activity is captured with parent/child IDs, timing, and step logs
Adjust on the Fly: Modify agent instructions, add/remove tools, or change process type without rebuilding

Troubleshooting & Optimization

Problem	Solution
Deadlocks (agents waiting indefinitely)	Ensure timeouts are set on feedback loops and waiting conditions
Worker failure or retry loops	Use Condition nodes or error-handling agents for recovery
Scaling bottlenecks	ParallelWorker patterns allow concurrent tasks with load balancing
Overlapping agent roles	Clarify role definitions and assign non-overlapping responsibilities
Poor result quality	Add review/editor agents, improve task descriptions, enable shared memory

Best Practices & Advanced Tips

Keep roles clear: Overlapping authorities dilute problem-solving effectiveness
Recover from errors: Supervisors can reassign or retry sub-tasks automatically
Limit depth and chains: Excessively deep agent hierarchies are hard to debug — use logs and Trace nodes
Integrate with Analytics: CrewAI runs can connect to LangFuse, Arize, or other observability tools for cross-session analysis
Use Parallel Patterns: Independent tasks should run concurrently to reduce overall execution time
Leverage Shared Memory: Enable for crews where agents need context from each other's work

Sequential Agents — Pipeline of roles, each building on the previous step
Hierarchical Delegation — Supervisor hands off goals to specialized workers
Voting System — Multiple agents generate outputs, best result chosen by consensus
Recursive Planning — Agents hand off sub-goals to other agents, including themselves

CrewAI unlocks complex automation, research pipelines, autonomous operations, and enterprise-level workflows — precisely coordinated, explainable, and scalable inside Forjinn.

CrewAI Integration — Full framework reference
Multi-Agent Systems — Architecture patterns
AutoGen — Alternative generative workflow framework
Google ADK — Google's agent development kit
Overview — Forjinn platform builder modes

CrewAI: How to Use CrewAI in Forjinn