Loading...
Loading...
Spawn specialist agents that collaborate on complex tasks — parallel research, code review pipelines, and multi-perspective analysis.
A team is a group of specialist AI agents coordinated by a supervisor. Each member has a defined role and goal. The supervisor assigns tasks, collects results, and synthesizes a final answer.
"Create a team called 'security-review' with these members: 1. Code Reviewer — find bugs and code quality issues 2. Security Analyst — check for vulnerabilities and attack vectors 3. Performance Expert — identify bottlenecks and optimization opportunities Run them in parallel on src/auth.ts"
| Mode | How It Works | Best For |
|---|---|---|
| Parallel | All agents work simultaneously on the same task. Results are merged. | Independent analysis, brainstorming, multi-perspective review |
| Sequential | Agents work one at a time. Each builds on the previous agent's output. | Pipeline workflows: research → plan → implement → review |
| Discussion | Agents respond to each other in rounds. The supervisor synthesizes consensus. | Architecture decisions, trade-off analysis, debate |
Group chats are multi-turn conversations between specialist agents. Unlike teams (which produce a single output), group chats let agents discuss back and forth, building on each other's responses.
Agents respond in order. Each gets a turn before cycling.
A random agent responds each turn. Good for diverse perspectives.
The supervisor picks the best agent to respond based on context.
Research agent explores the codebase and requirements. Code agent implements the solution. Review agent checks for bugs, security issues, and style violations. Sequential execution.
Multiple agents search different sources simultaneously — one checks documentation, another searches the web, a third analyzes existing code. Results are merged into a comprehensive answer.
Three agents debate the best approach: one advocates for simplicity, one for scalability, one for developer experience. The supervisor synthesizes a balanced recommendation.
Security analyst, performance expert, and code quality reviewer all examine the same PR in parallel. Each catches different issues.