Orchestration Basics (Config-first)

Goal: Learn how to switch between orchestration modes to control agent interactions.

Prerequisites

Complete first-agent.md to understand basic project structure.

What is Orchestration?

Orchestration controls how multiple agents work together:

• route: Single agent handles each request
• collaborative: Agents work in parallel and combine results
• sequential: Agents work in a specific order
• loop: Single agent iterates until condition is met
• mixed: Combination of collaborative and sequential

1) Create a multi-agent project

agentcli create orchestration-demo --template research-assistant
Set-Location orchestration-demo

This creates a project with multiple agents: researcher, analyzer, synthesizer.

2) Try collaborative mode (default)

Check your agentflow.toml:

[orchestration]
mode = "collaborative"
timeout_seconds = 30
collaborative_agents = ["researcher", "analyzer", "synthesizer"]

Run the project:

agentcli validate
go run . -m "Research the benefits of electric vehicles"

Expected: All agents process the query in parallel and combine their results.

3) Switch to sequential mode

Edit agentflow.toml:

[orchestration]
mode = "sequential"
sequential_agents = ["researcher", "analyzer", "synthesizer"]

Run again:

agentcli validate
go run . -m "Research the benefits of electric vehicles"

Expected: Researcher runs first, then analyzer processes researcher's output, then synthesizer creates final result.

4) Try route mode

Edit agentflow.toml:

[orchestration]
mode = "route"

Run again:

agentcli validate
go run . -m "Research the benefits of electric vehicles"

Expected: System routes to the most appropriate single agent based on the request.

5) Inspect execution (optional)

If available in your CLI version:

agentcli trace --help

Key Differences

• Collaborative: Fast, parallel processing, best for diverse perspectives
• Sequential: Structured pipeline, each agent builds on previous results
• Route: Single expert agent, fastest execution, good for specialized tasks
• Loop: Iterative refinement, good for improving quality over multiple passes

Next Steps

• Add memory to agents: memory-basics.md
• Enable tools: tools-basics.md

Verification checklist

• [ ] agentcli create succeeded
• [ ] Tried at least 2 different orchestration modes
• [ ] agentcli validate passed for each mode
• [ ] Observed different behavior between modes