Getting Started with Multi-Agent Development: A Practical Tutorial

Jun 6, 2026

What You'll Build

In this tutorial, you'll create a multi-agent code review system using Doforu. The system will:

  1. Analyze code across multiple files simultaneously
  2. Assign specialized reviewers (security, performance, style)
  3. Aggregate results into a unified report
  4. Apply suggested fixes automatically

Prerequisites

  • Doforu installed (install guide)
  • Basic understanding of terminal/CLI
  • An API key for your preferred LLM provider

Step 1: Understanding the Agent Orchestration Model

Before writing code, let's understand Doforu's core abstraction. A multi-agent workflow in Doforu consists of:

  • Main Agent: The orchestrator that receives your goal and decomposes it
  • Sub-Agents: Specialized agents that execute individual tasks
  • Task Graph: A DAG (directed acyclic graph) defining task dependencies
  • Tools: Capabilities each agent can invoke (read files, search, execute commands)

Step 2: Creating Your First Multi-Agent Configuration

Create a file called review-workflow.json:

{
  "name": "code-review-workflow",
  "agents": [
    {
      "name": "security-reviewer",
      "role": "Identify security vulnerabilities including XSS, SQL injection, and insecure dependencies",
      "tools": ["read_file", "grep", "web_search"],
      "model": "claude-sonnet-4-2026"
    },
    {
      "name": "performance-reviewer",
      "role": "Identify performance bottlenecks, N+1 queries, memory leaks, and optimization opportunities",
      "tools": ["read_file", "shell"],
      "model": "gpt-5-2026"
    },
    {
      "name": "style-reviewer",
      "role": "Check code style, naming conventions, and adherence to project standards",
      "tools": ["read_file", "grep"],
      "model": "claude-sonnet-4-2026"
    }
  ],
  "workflow": {
    "parallel": ["security-reviewer", "performance-reviewer", "style-reviewer"],
    "aggregator": {
      "agent": "review-aggregator",
      "depends_on": ["security-reviewer", "performance-reviewer", "style-reviewer"],
      "role": "Merge all review results into a single structured report with severity levels"
    }
  }
}

Step 3: Running the Multi-Agent Workflow

Execute the workflow with Doforu's CLI:

# Run the multi-agent code review
doforu run review-workflow.json --target ./src

# Or use shorthand for ad-hoc multi-agent tasks
doforu review ./src --agents=3 --focus=security,performance,style

Doforu will immediately spawn 3 sub-agents in parallel, each analyzing your codebase from their specialized perspective.

Step 4: Understanding Parallel Execution

This is where Doforu's power shines. Instead of running reviews sequentially:

Sequential (traditional tools):
  Step 1: Security Review (30s) → Step 2: Performance Review (30s) → Step 3: Style Review (30s)
  Total: ~90 seconds

Parallel (Doforu):
  Step 1: All 3 reviews start simultaneously
  Total: ~30 seconds (3x faster!)

For larger workflows with 10+ agents, the speedup becomes dramatic.

Step 5: Real-Time Monitoring

Doforu provides real-time visibility into agent progress:

# Watch agent execution in real-time
doforu run review-workflow.json --watch

# Output shows:
# [security-reviewer]    🔍 Scanning for XSS vulnerabilities... DONE
# [performance-reviewer] ⚡ Analyzing query patterns... IN PROGRESS
# [style-reviewer]       📐 Checking naming conventions... WAITING

Step 6: Consuming the Aggregated Result

Once all agents complete, the aggregator agent compiles the results:

{
  "summary": {
    "critical": 2,
    "high": 5,
    "medium": 12,
    "low": 23
  },
  "findings": [
    {
      "severity": "critical",
      "agent": "security-reviewer",
      "file": "src/auth/login.ts",
      "issue": "SQL injection vulnerability in raw query",
      "suggestion": "Use parameterized queries instead of string concatenation"
    }
  ],
  "auto_fixes_applied": 8
}

Advanced: Custom DAG Workflows

For complex scenarios, define dependencies between agents:

{
  "workflow": {
    "stages": [
      {
        "name": "analysis",
        "parallel": ["dependency-checker", "complexity-analyzer"]
      },
      {
        "name": "review",
        "parallel": ["security-reviewer", "performance-reviewer"],
        "depends_on": ["analysis"]
      },
      {
        "name": "report",
        "agent": "report-generator",
        "depends_on": ["review"]
      }
    ]
  }
}

This creates a pipeline where analysis completes before review begins, but agents within each stage run in parallel.

Next Steps

Ready to experience true multi-agent development? Install Doforu and start building.

Doforu Team

Doforu Team