Team Configuration¶

Teams and sequential chains in LLMling-agent allow you to create sophisticated message flows by composing message nodes. Any node (Agent, Team, or TeamRun) can be part of a team or chain, allowing for arbitrarily complex nested structures.

Example Configuration¶

agents:
  analyzer:
    model: openai:gpt-4
    system_prompts: ["You analyze code for issues."]

  planner:
    model: openai:gpt-4
    system_prompts: ["You create fix plans."]

  executor:
    model: openai:gpt-4
    system_prompts: ["You implement fixes."]

teams:
  analysis_team:
    mode: parallel  # Run members in parallel
    members:
      - analyzer
      - planner
    shared_prompt: "Focus on performance issues."  # Added to all members' prompts
    connections:
      - type: node
        name: executor
        wait_for_completion: true

  review_chain:
    mode: sequential  # Run members in sequence
    members:
      - analysis_team  # Teams can be members
      - executor
    shared_prompt: "This is a critical production system."

  monitoring:
    mode: parallel
    members:
      - review_chain  # Chains can be members
      - performance_monitor
    connections:
      - type: file
        path: "logs/team_output.txt"

Components¶

Mode¶

parallel: Members execute simultaneously (like & operator)
sequential: Members execute in sequence (like | operator)

Members¶

References to agents or other teams

Can include any message node type:

members:
  - single_agent  # Individual agent
  - parallel_team # Team
  - processing_chain  # TeamRun

Shared Prompt¶

Additional context provided to all team members:

teams:
  security_team:
    mode: parallel
    members: [analyzer, validator]
    shared_prompt: "Focus on security implications."  # Added to each member's input

Connections¶

Message forwarding configuration:

connections:
  - type: node
    name: final_reviewer
    queued: true
    queue_strategy: latest
    wait_for_completion: true
    filter_condition:
      type: word_match
      words: ["critical", "urgent"]

Nesting Capabilities¶

Teams can be arbitrarily nested to create complex workflows:

teams:
  analysis:
    mode: parallel
    members: [analyzer, planner]

  execution:
    mode: sequential
    members: [validator, executor]

  workflow:
    mode: sequential
    members:
      - analysis  # Parallel team
      - execution  # Sequential chain
    connections:
      - type: node
        name: final_review

Connection Control¶

Message Flow¶

wait_for_completion: Whether to wait for target to complete
queued: Queue messages for manual processing
queue_strategy: How to handle queued messages (latest/concat/buffer)

Filtering¶

filter_condition:
  type: word_match
  words: ["urgent", "critical"]
  case_sensitive: false

Transformation¶

transform: myapp.transforms.process_message  # Python callable

Monitoring¶

All teams provide statistics: - Message count - Token usage - Execution timing - Cost tracking

Example Complex Workflow¶

teams:
  # Initial analysis (parallel)
  analysis:
    mode: parallel
    members: [code_analyzer, security_checker]
    shared_prompt: "Focus on production impact."

  # Planning chain (sequential)
  planning:
    mode: sequential
    members: [issue_classifier, fix_planner]
    shared_prompt: "Consider dependencies."

  # Execution group (parallel)
  execution:
    mode: parallel
    members: [code_fixer, test_runner]

  # Complete workflow
  full_pipeline:
    mode: sequential
    members:
      - analysis  # Parallel analysis
      - planning  # Sequential planning
      - execution  # Parallel execution
    connections:
      - type: node
        name: final_reviewer
        wait_for_completion: true
      - type: file
        path: "reports/{date}_workflow.txt"

This configuration creates a sophisticated workflow where: 1. Analysis runs in parallel 2. Planning happens sequentially 3. Execution runs in parallel again 4. Results are reviewed and logged

All while maintaining type safety and providing comprehensive monitoring.