A lightweight, generic graph-based workflow engine for Go, inspired by LangGraph. This SDK allows you to build stateful, multi-step workflows with cycles, conditional branching, and human-in-the-loop capabilities (via execution control).

• 🕸 Graph-based Workflow: Define nodes and edges to create complex execution flows.
• 🔀 Conditional Branching: Dynamically route execution based on state.
• 🔁 Cycles Support: Built-in loop detection and iteration limits.
• 💾 State Management: Strongly typed generic state passing between nodes.
• 🛑 Early Termination: Control execution flow explicitly from nodes.
• 📡 Streaming Callbacks: Hook into execution for logging or real-time updates.

go get github.com/restayway/langgraph-go

Here's a simple example of how to define and run a graph:

package main import ( "context" "fmt" "github.com/restayway/langgraph-go" ) // 1. Define your Custom State type MyState struct { *langgraph.DefaultState Data string } func main() { // 2. Create the Graph g := langgraph.NewGraph[*MyState]() // 3. Add Nodes g.AddNode("start", func(ctx context.Context, s *MyState) (*MyState, error) { fmt.Println("Starting processing:", s.Data) s.Data += "_processed" return s, nil }) // 4. Define Edges g.SetEntryPoint("start") g.AddEdge("start", langgraph.END) // 5. Compile app, _ := g.Compile() // 6. Run final, _ := app.Run(context.Background(), &MyState{ DefaultState: &langgraph.DefaultState{}, Data: "input", }) fmt.Println(final.Data) // Output: input_processed }

Check the examples/ directory for more complex use cases like conditional routing (chatbots).

A node is simply a function that takes the current state and returns a verified state.

type NodeFunc[S State] func(ctx context.Context, state S) (S, error)

Edges define the transition between nodes.

• AddEdge(from, to): Direct transition.
• AddConditionalEdge(from, conditionFunc): Dynamic transition.

Your state struct must implement the State interface to allow the engine to control flow (e.g., stop execution).

type State interface { SetShouldEnd(bool) GetShouldEnd() bool }

You can embed langgraph.DefaultState to handle this automatically.