> ## Documentation Index
> Fetch the complete documentation index at: https://overmind-b84ae13c-codex-workshop-queue-capacity.mintlify.site/llms.txt
> Use this file to discover all available pages before exploring further.

# TypeScript SDK

> There is no TypeScript tracing SDK; instrument Node.js with OpenTelemetry against the OTLP endpoint.

<Warning>
  There is no TypeScript SDK. Instrument Node.js with OpenTelemetry directly.
</Warning>

Node.js services trace like any other OpenTelemetry client:

* Export OTLP/HTTP protobuf to the [OTLP endpoint](/core/observability#the-otlp-endpoint) with `x-api-key` in the exporter headers.
* Stamp `overmind.capability.id` as a resource attribute — the capability's UUID or its slug from `overmind.toml` — so spans bind to the capability. `overmind.capability.name` is an optional display label and never binds.
* Mark the run boundary with `overmind.span.type = "entry_point"` and `overmind.unit_kind = "run"` on the root span, and the result span with `overmind.delivery = "true"`, so [trace scoring](/agent-testing/trace-scoring) has a unit and a terminal to judge.
* Put `conversation.id` on every span of a multi-turn exchange to group them into a session.
* To bind units to scanned behaviours, stamp `code.namespace` and `code.function.name` on the spans that wrap anchor functions, and `vcs.ref.head.revision` on the resource.

The [attribute table](/core/observability#the-attributes-overmind-reads) lists every key the pipeline reads, including the OTel GenAI and OpenLLMetry conventions that the OpenAI and Anthropic instrumentations for Node already emit.


This documentation is built and hosted on [Mintlify](https://mintlify.com), a developer documentation platform.