Summary
Cartesia publishes an official TypeScript/JavaScript SDK, @cartesia/cartesia-js on npm (Stainless-generated, currently v3.x), providing execution APIs for real-time and batch text-to-speech generation using its Sonic model family. This repository has zero instrumentation for any Cartesia SDK surface — no wrapper, no diagnostics channel, no plugin, no auto-instrumentation config, no e2e scenario.
What's missing
Execution surfaces exposed by @cartesia/cartesia-js that are not instrumented:
| SDK Method |
Description |
client.tts.generate({ modelId, transcript, voice, outputFormat }) |
Batch text-to-speech generation, returns synthesized audio bytes |
client.tts.sse({ modelId, transcript, ... }) |
Server-Sent-Events streaming text-to-speech generation |
client.tts.websocket(...) / ws.send({ transcript, ... }) |
Low-latency WebSocket streaming TTS generation, the SDK's primary realtime execution path |
client.voices.clone(...) / client.voices.localize(...) |
Voice generation/transformation execution surfaces |
The SDK ships its own REST/WebSocket client (ws as a peer dependency in Node.js) — it does not depend on openai or the Vercel ai package, so none of it is picked up by existing OpenAI/AI-SDK auto-instrumentation.
No coverage in any instrumentation layer:
- No wrapper function (e.g.
wrapCartesia()) in js/src/wrappers/ (52 files checked)
- No channel definitions or plugin in
js/src/instrumentation/plugins/ (108 files checked)
- No auto-instrumentation config in
js/src/auto-instrumentations/configs/ (34 files checked, including all.ts)
- No e2e test scenario under
e2e/scenarios/
- Repo-wide case-insensitive grep for
cartesia returns zero matches anywhere in the repo
Weekly downloads
Weekly downloads: 60,223 (as of 2026-09-07; window 2026-08-31 to 2026-09-06; https://api.npmjs.org/downloads/point/last-week/@cartesia/cartesia-js)
Braintrust docs status
not_found. Cartesia does not appear anywhere on the Braintrust tracing-integrations index — https://www.braintrust.dev/docs/guides/tracing/integrations — which enumerates Tracing integrations (OpenTelemetry, Vercel AI SDK, LangChain, LlamaIndex, Cloudflare Workers AI, etc.), Agent frameworks (LiveKit Agents, Mastra, Pipecat, etc.), Eval frameworks, and Testing frameworks. Cartesia is also absent from https://www.braintrust.dev/docs/integrations. The only path to Braintrust tracing for Cartesia calls made through the Vercel AI SDK (a different, generic provider) is unrelated to Cartesia's own native SDK, which this issue is about.
Upstream references
Braintrust docs sources checked
Local files inspected
- Repo-wide case-insensitive grep for
cartesia — zero matches
js/src/wrappers/*.ts — full directory listing (52 files) — no cartesia.ts wrapper
js/src/instrumentation/plugins/*.ts — full directory listing (108 files) — no Cartesia channels or plugin file
js/src/auto-instrumentations/configs/*.ts — full directory listing (34 files, including all.ts) — no Cartesia config entry
js/src/instrumentation/plugins/instrumentation-names.test.ts — canonical INSTRUMENTATION_NAMES list has no Cartesia entry
e2e/scenarios/ — no Cartesia test scenario
Summary
Cartesia publishes an official TypeScript/JavaScript SDK,
@cartesia/cartesia-json npm (Stainless-generated, currently v3.x), providing execution APIs for real-time and batch text-to-speech generation using its Sonic model family. This repository has zero instrumentation for any Cartesia SDK surface — no wrapper, no diagnostics channel, no plugin, no auto-instrumentation config, no e2e scenario.What's missing
Execution surfaces exposed by
@cartesia/cartesia-jsthat are not instrumented:client.tts.generate({ modelId, transcript, voice, outputFormat })client.tts.sse({ modelId, transcript, ... })client.tts.websocket(...)/ws.send({ transcript, ... })client.voices.clone(...)/client.voices.localize(...)The SDK ships its own REST/WebSocket client (
wsas a peer dependency in Node.js) — it does not depend onopenaior the Vercelaipackage, so none of it is picked up by existing OpenAI/AI-SDK auto-instrumentation.No coverage in any instrumentation layer:
wrapCartesia()) injs/src/wrappers/(52 files checked)js/src/instrumentation/plugins/(108 files checked)js/src/auto-instrumentations/configs/(34 files checked, includingall.ts)e2e/scenarios/cartesiareturns zero matches anywhere in the repoWeekly downloads
Weekly downloads: 60,223 (as of 2026-09-07; window 2026-08-31 to 2026-09-06; https://api.npmjs.org/downloads/point/last-week/@cartesia/cartesia-js)
Braintrust docs status
not_found. Cartesia does not appear anywhere on the Braintrust tracing-integrations index — https://www.braintrust.dev/docs/guides/tracing/integrations — which enumerates Tracing integrations (OpenTelemetry, Vercel AI SDK, LangChain, LlamaIndex, Cloudflare Workers AI, etc.), Agent frameworks (LiveKit Agents, Mastra, Pipecat, etc.), Eval frameworks, and Testing frameworks. Cartesia is also absent from https://www.braintrust.dev/docs/integrations. The only path to Braintrust tracing for Cartesia calls made through the Vercel AI SDK (a different, generic provider) is unrelated to Cartesia's own native SDK, which this issue is about.Upstream references
Braintrust docs sources checked
Local files inspected
cartesia— zero matchesjs/src/wrappers/*.ts— full directory listing (52 files) — nocartesia.tswrapperjs/src/instrumentation/plugins/*.ts— full directory listing (108 files) — no Cartesia channels or plugin filejs/src/auto-instrumentations/configs/*.ts— full directory listing (34 files, includingall.ts) — no Cartesia config entryjs/src/instrumentation/plugins/instrumentation-names.test.ts— canonicalINSTRUMENTATION_NAMESlist has no Cartesia entrye2e/scenarios/— no Cartesia test scenario