Teams

Overview

Teams enable parallel agent execution. Instead of running steps sequentially (like chains), teams spawn multiple agents that work on sub-tasks simultaneously and combine results.

This is useful when different aspects of a financial analysis can be computed independently:

• One agent analyzes income trends while another reviews expenses
• One agent searches for subscriptions while another categorizes transactions
• Multiple agents process different time periods in parallel

When to Use Teams vs Chains

Use Case	Chains	Teams
Sequential dependencies (step B needs step A’s output)	Best	Not ideal
Independent sub-tasks that can run in parallel	Slow	Best
Simple 2-4 step pipelines	Best	Overkill
Complex analysis with multiple dimensions	OK	Best
Reproducible, deterministic output	Best	Less predictable

Team Format

Teams are defined in YAML with a list of agents:

name: comprehensive-review
description: Full financial review with parallel analysis
agents:
  - id: income-analyst
    systemPrompt: Analyze all income sources, trends, and projections.
    tools: [transaction_search, spending_summary]
  - id: expense-analyst
    systemPrompt: Analyze spending patterns, categories, and anomalies.
    tools: [transaction_search, spending_summary, anomaly_detect]
  - id: subscription-auditor
    systemPrompt: Find all recurring charges and flag potential savings.
    tools: [transaction_search, anomaly_detect]
synthesis:
  systemPrompt: >
    Combine the results from all agents into a comprehensive financial
    review report with income analysis, expense breakdown, and subscription
    audit findings.

Execution

1. All agents start simultaneously with their respective tools and prompts
2. Each agent works independently, producing its own results
3. When all agents complete, the synthesis step combines their outputs
4. The final synthesized report is presented to the user

Performance

Teams can significantly reduce execution time for complex analyses. A 4-agent team running in parallel completes in roughly the time of the slowest individual agent, rather than the sum of all agents.