Langtrace Tutorial: Get Started in 5 Minutes [2026]

What it does:

All Langtrace spans conform to the emerging OpenTelemetry GenAI semantic conventions, so prompts, completions, token counts, model parameters, tool calls, and retrieval results are stored in standardized attributes. This means traces can be exported via OTLP to Grafana Tempo, Datadog, Signoz, Jaeger, or any compliant backend without transformation, giving teams full portability over their telemetry data and avoiding vendor lock-in.

Use case:

What it does:

Initialization takes two lines: import the SDK and call init with an API key. Every supported LLM, framework, and vector DB call is then traced automatically with full prompt content, completion text, token counts, latency, and cost — no manual span creation required. The Python SDK supports OpenAI, Anthropic, Gemini, Cohere, Groq, Mistral, Perplexity, Ollama, LangChain, LlamaIndex, CrewAI, DSPy, AutoGen, Pinecone, Chroma, Weaviate, and Qdrant. The TypeScript SDK covers a similar set of providers and frameworks.

Use case:

What it does:

Aggregated dashboards display cost per model, user, project, prompt template, and time range, alongside p50/p95/p99 latency for individual operations and full traces. Cost is calculated automatically using each provider's published token pricing. Teams use these dashboards to set budget alerts, identify cost spikes from specific features or tenants, and present attribution data to finance stakeholders for AI infrastructure spend.

Use case:

What it does:

Saved prompts can be versioned, edited, and tested across multiple models in a side-by-side playground. Experiment results are persisted so teams can compare output quality, latency, and cost across model versions and prompt variations before deploying changes to production. This workflow supports systematic prompt engineering rather than ad-hoc testing in notebooks.

Use case:

What it does:

Any production trace can be added to a dataset, labeled by human annotators, and run through built-in or custom evaluators measuring accuracy, faithfulness, toxicity, JSON schema compliance, and other quality metrics. Custom evaluator functions can be defined in Python for domain-specific scoring. This creates a feedback loop where production issues are captured, annotated, evaluated, and used to validate fixes before redeployment.

Use case:

What it does:

A single Docker Compose file launches the server, Postgres for metadata, and ClickHouse for high-performance trace storage. Kubernetes Helm charts are available for production deployments that require horizontal scaling. Self-hosted instances receive all features available in the managed Cloud offering, with the only trade-off being that teams manage their own infrastructure, upgrades, and backups.

Use case:

What it does:

Workspaces, projects, role-based access control, and API key scoping let larger organizations separate staging from production traffic and limit which team members can access sensitive trace data. This is essential for enterprise deployments where multiple teams share a single Langtrace instance but need isolation between their observability data and configurations.

Use case: