workflow-parallel

README

Parallel Workflow Example

This example demonstrates how to use parallel workflow agents to execute multiple agents concurrently.

Overview

The parallel workflow agent executes multiple agents at the same time, with various options for controlling concurrency and merging results. This is useful for:

• Getting multiple perspectives on the same topic
• Racing agents to get the fastest response
• Distributing work across multiple specialized agents
• Gathering data from multiple sources simultaneously

Features Demonstrated

1. Concurrent Execution: Multiple agents run simultaneously
2. Concurrency Control: Limit the number of agents running at once
3. Merge Strategies:
   • MergeAll: Combine all results into one state
   • MergeFirst: Use the first completed result
   • MergeCustom: Apply custom logic to merge results
4. Timeout Support: Set maximum time for workflow completion
5. Error Handling: Configure how to handle partial failures

Usage

# Run with mock agents
go run main.go

# Run with actual LLM providers (requires API keys)
export ANTHROPIC_API_KEY=your-key
export OPENAI_API_KEY=your-key
go run main.go

Examples

1. Parallel Analysis

Analyzes a topic from multiple perspectives simultaneously:

• Technical analyst (Claude)
• Business analyst (GPT-4)
• Ethical analyst (Claude)

All three analyses run concurrently and results are merged.

2. Racing Agents

Multiple agents race to provide the fastest response:

• Fast agent (100ms)
• Medium agent (300ms)
• Slow agent (500ms)

With MergeFirst strategy, the workflow completes as soon as the first agent finishes.

3. Custom Merge

Demonstrates custom merge logic by calculating average scores from multiple scoring agents.

Code Structure

// Create parallel workflow
workflow := workflow.NewParallelAgent("analysis").
    WithMaxConcurrency(3).
    WithMergeStrategy(workflow.MergeAll).
    AddAgent(agent1).
    AddAgent(agent2).
    AddAgent(agent3)

// Run workflow
result, err := workflow.Run(ctx, initialState)

Configuration Options

• WithMaxConcurrency(int): Limit concurrent executions
• WithMergeStrategy(MergeStrategy): How to combine results
• WithMergeFunc(func): Custom merge logic
• WithTimeout(duration): Maximum execution time

Merge Strategies

MergeAll (Default)

All agent results are stored in parallel_results map:

results := result.Get("parallel_results")
// results["agent1"] contains agent1's output
// results["agent2"] contains agent2's output

MergeFirst

Returns the state from the first agent to complete successfully.

MergeCustom

Apply custom logic to combine results:

workflow.WithMergeFunc(func(results map[string]*domain.State) *domain.State {
    // Custom merge logic
    merged := domain.NewState()
    // Process results...
    return merged
})

Next Steps

See the conditional workflow example for branching logic, or the loop workflow example for iterative processing.