Flyt: A Zero-Dependency Workflow Framework Brings Norwegian Simplicity to Go

In a landscape dominated by orchestration behemoths, Flyt offers Go developers a refreshingly minimalist approach to workflow management. Pronounced "fleet," this new open-source framework strips away dependencies while delivering sophisticated capabilities for building resilient, composable processes—from simple linear tasks to complex branching workflows.

The Anatomy of Flow

At Flyt's core lies a simple yet powerful triad:

1. Nodes: Building blocks with three-phase execution:

node := flyt.NewNode(
    flyt.WithPrepFunc(func(ctx context.Context, shared *flyt.SharedStore) (any, error) {
        data, _ := shared.Get("input")
        return data, nil
    }),
    flyt.WithExecFunc(func(ctx context.Context, prepResult any) (any, error) {
        return transform(prepResult), nil
    }),
)

1. Actions: Dynamic routing between nodes via string-based decisions (e.g., "success", "retry", "fallback")

2. Shared Store: Thread-safe data passing with merge capabilities:

shared.Merge(map[string]any{
    "user_id": 123,
    "config": map[string]any{"timeout": 30},
})

Resilience by Design

Flyt bakes in production-grade patterns:

• Automatic Retries: Configurable backoff strategies
• Fallback Mechanisms: Graceful degradation via the FallbackNode interface
• Rate Limiting: Custom node implementations with token buckets

Scaling Patterns

For complex workloads, Flyt offers:

// Concurrent batch processing
batchNode := flyt.NewBatchNode(processFunc, true)

// Nested flows as composable units
validationFlow := createValidationFlow()
mainFlow.Connect(fetchNode, "validate", validationFlow)

// Worker pools for custom concurrency
pool := flyt.NewWorkerPool(10)
pool.Submit(func() { /* ... */ })

Why Minimalism Matters

In contrast to dependency-heavy alternatives, Flyt's zero-external-dependency philosophy reduces:

• Security vulnerabilities
• Version conflicts
• Binary bloat

This makes it ideal for embedded systems, lightweight microservices, and environments with strict compliance requirements.

Real-World Implementations

The framework shines in AI/ML pipelines, showcased in included examples:

• LLM Streaming: Real-time OpenAI responses with Server-Sent Events
• Agent Systems: AI agents with web search capabilities
• Distributed Tracing: Langfuse integration for workflow observability

As distributed systems grow increasingly complex, Flyt offers a compelling blend of simplicity and power—proving that sometimes, the most effective flows come from the lightest currents.

Source: Flyt GitHub Repository