LangSmith Observability Provider

The LangSmith provider exports LLM call data to LangSmith, LangChain’s platform for debugging, testing, and monitoring LLM applications. It implements the o11y.TraceExporter interface and sends run data through the LangSmith batch API.

Choose LangSmith when you want a managed platform with strong debugging and testing capabilities for LLM applications. LangSmith organizes traces into projects and provides detailed run inspection with inputs, outputs, and token usage. For an open-source self-hostable alternative, consider Langfuse. For workspace-level analytics, consider Opik.

Installation

go get github.com/lookatitude/beluga-ai/o11y/providers/langsmith

Configuration

Option	Type	Default	Description
WithAPIKey(key)	string	(required)	LangSmith API key (lsv2_...)
WithProject(name)	string	"default"	LangSmith project name
WithBaseURL(url)	string	https://api.smith.langchain.com	API endpoint
WithTimeout(d)	time.Duration	10s	HTTP request timeout

Environment variables:

Variable	Maps to
LANGSMITH_API_KEY	WithAPIKey

Basic Usage

package main

import (
    "context"
    "log"
    "os"
    "time"

    "github.com/lookatitude/beluga-ai/o11y"
    "github.com/lookatitude/beluga-ai/o11y/providers/langsmith"
)

func main() {
    exporter, err := langsmith.New(
        langsmith.WithAPIKey(os.Getenv("LANGSMITH_API_KEY")),
        langsmith.WithProject("my-project"),
    )
    if err != nil {
        log.Fatal(err)
    }

    err = exporter.ExportLLMCall(context.Background(), o11y.LLMCallData{
        Model:        "gpt-4o",
        Provider:     "openai",
        InputTokens:  500,
        OutputTokens: 150,
        Duration:     1200 * time.Millisecond,
        Cost:         0.003,
        Response:     "The capital of France is Paris.",
        Metadata: map[string]any{
            "temperature": 0.7,
            "max_tokens":  2048,
        },
    })
    if err != nil {
        log.Fatal(err)
    }
}

Project Organization

LangSmith organizes traces into projects. Use the WithProject option to route traces to a specific project:

// Development traces
devExporter, err := langsmith.New(
    langsmith.WithAPIKey(os.Getenv("LANGSMITH_API_KEY")),
    langsmith.WithProject("my-app-dev"),
)

// Production traces
prodExporter, err := langsmith.New(
    langsmith.WithAPIKey(os.Getenv("LANGSMITH_API_KEY")),
    langsmith.WithProject("my-app-prod"),
)

Run Model

Each ExportLLMCall creates a LangSmith “run” with:

• Run type: "llm"
• Name: Formatted as "provider/model" (e.g., "openai/gpt-4o")
• Inputs: The message history sent to the model
• Outputs: The model’s response
• Extras: Token counts, cost, duration, and any additional metadata
• Session name: The configured project name

With MultiExporter

Combine LangSmith with other observability providers:

import (
    "github.com/lookatitude/beluga-ai/o11y"
    "github.com/lookatitude/beluga-ai/o11y/providers/langsmith"
    "github.com/lookatitude/beluga-ai/o11y/providers/langfuse"
)

lsExporter, err := langsmith.New(
    langsmith.WithAPIKey(os.Getenv("LANGSMITH_API_KEY")),
)
if err != nil {
    log.Fatal(err)
}

lfExporter, err := langfuse.New(
    langfuse.WithPublicKey(os.Getenv("LANGFUSE_PUBLIC_KEY")),
    langfuse.WithSecretKey(os.Getenv("LANGFUSE_SECRET_KEY")),
)
if err != nil {
    log.Fatal(err)
}

multi := o11y.NewMultiExporter(lsExporter, lfExporter)

Authentication

LangSmith uses API key authentication via the x-api-key HTTP header. API keys are prefixed with lsv2_.

Error Handling

err = exporter.ExportLLMCall(ctx, data)
if err != nil {
    // Errors include authentication failures, network issues, and API errors
    log.Printf("LangSmith export failed: %v", err)
}

The Flush method is a no-op since the provider sends data synchronously via HTTP.