LANGCHAINGO.md

Using langgraph-go with langchaingo

langchaingo is a popular Go toolkit for models, embeddings, and document pipelines. langgraph-go does not replace it. Use langchaingo (or any provider SDK) for model I/O, and langgraph-go for durable graph execution, interrupts, and multi-agent orchestration.

Recommended split

Concern	Prefer
Chat completions, embeddings, loaders, vector stores	langchaingo
Typed graph, checkpoint, resume, fan-out, supervisors	langgraph-go
Tool allowlists / human approval	prebuilt.GuardTools
Long-term memory in the graph	langgraph-go store + optional retrieval

Message bridge pattern

langgraph-go agents speak prebuilt.Message / AgentState. Adapters map provider messages into that shape. A typical langchaingo bridge:

// Pseudo-code: adapt langchaingo llms.MessageContent into prebuilt.AgentState.
// Keep this adapter in your application module so core langgraph-go stays free
// of a hard langchaingo dependency.

import (
    "context"

    "github.com/tmc/langchaingo/llms"
    "github.com/ybszm/langgraph-go/graph"
    "github.com/ybszm/langgraph-go/prebuilt"
)

type LangChainGoModel struct {
    Model llms.Model
}

func (m LangChainGoModel) Invoke(
    ctx context.Context,
    state prebuilt.AgentState,
    _ graph.Runtime,
) (prebuilt.AssistantMessage, error) {
    messages, err := state.ProviderMessages()
    if err != nil {
        return prebuilt.AssistantMessage{}, err
    }
    // Convert prebuilt.Message → []llms.MessageContent in your app.
    content, err := toLangChainMessages(messages)
    if err != nil {
        return prebuilt.AssistantMessage{}, err
    }
    resp, err := m.Model.GenerateContent(ctx, content)
    if err != nil {
        return prebuilt.AssistantMessage{}, err
    }
    return fromLangChainResponse(resp)
}

Then:

agent, err := prebuilt.NewQuickAgent(prebuilt.QuickAgentConfig{
    Model: LangChainGoModel{Model: openaiModel},
    Tools: tools,
})

Tools

• Prefer prebuilt.Tool / ToolFunc for graph-native tools.
• Wrap external HTTP tools with timeouts in the Run closure.
• Apply prebuilt.GuardTools before binding tools to the agent.

What not to do

• Do not import langchaingo from langgraph-go core packages (keeps the module lean).
• Do not expect Python LangGraph + Python LangChain APIs to map 1:1.

Minimal checklist

1. Implement prebuilt.ChatModel[prebuilt.AgentState] over your langchaingo model.
2. Build with NewQuickAgent or NewChatModelAgent.
3. Add graph.WithPersistence when you need resume.
4. Stream with AgentRunner.Query for application UIs.