Plug into your stack. Trace, evaluate, ship.
25+ first-class integrations. Python and TypeScript SDKs, every major model provider, every agent framework worth using, OpenTelemetry-native end to end.
Languages
Wrap your agent in one line. Python and TypeScript, both OpenTelemetry-native.
OpenTelemetry & REST
Send spans from any language. Already on OTEL? Point the exporter at LangWatch.
LLMs & model gateways
Trace any provider. The AI Gateway gives you one endpoint, virtual keys, fallback, and budgets across all of them.
GPT-5, GPT-4.1, o-series. Auto-instrumented end to end.
Claude Opus, Sonnet, Haiku. Tool use and streaming captured.
Gemini and Vertex AI traces with prompts, tools, and usage.
Bedrock Runtime captured with model, region, and token usage.
Azure-deployed OpenAI models with deployment metadata.
Google Cloud Vertex AI for Gemini and partner models.
Ultra-low-latency inference, traced like any other provider.
Local models traced through the OpenAI-compatible adapter.
Hit 100+ providers behind one OpenAI-compatible interface.
Frameworks & orchestrators
Drop LangWatch into whatever you build agents with. Auto-instrumentation captures every step, tool call, and handoff.
The official Agents SDK, traced step by step.
Graph-state agents with every node, edge, and retry captured.
Chains, runnables, and callbacks instrumented automatically.
Multi-agent crews with role, task, and tool-call visibility.
Type-safe agents with structured I/O and tool traces.
High-performance Python framework with native tracing.
TypeScript-first agent framework with full OTEL bridge.
streamText, generateText, and tool calls traced end to end.
The Agent Development Kit, instrumented for production.
Visual flows pushed through LangWatch with one toggle.
No-code agent builder with the LangWatch analytics tile.
Prompt optimization
Run real optimizers against your traces and datasets — not vibes.
Same three lines. Whatever your stack.
Install the SDK, set one environment variable, wrap your agent. Every framework on this page lands the same way: traces, evals, and prompt history in the platform without touching your call sites.
Five-minute quickstartimport langwatch
@langwatch.trace()
def run_agent(query: str) -> str:
# your agent, untouched
...Don't see your stack? It probably works anyway.
If it speaks OpenTelemetry, it speaks LangWatch. Point your exporter at our OTLP endpoint, or open a ticket and we'll wire it up with you.