Audra SDKs: npm audra and pip install audra — Audra

There are two official SDKs — npm i audra for Node, pip install audra for Python — plus the zero-install path: the OpenAI SDK pointed at the compat base URL. They are not redundant. The native SDKs speak the v2 wire format (text, format, Idempotency-Key, the jobs API); the OpenAI SDK path is the migration shortcut. This post is the decision guide and the install snippet for each.

npm: the Node client

npm install audra

import { Audra } from "audra";
import { writeFile } from "node:fs/promises";

const audra = new Audra({ apiKey: process.env.AUDRA_API_KEY });

const job = await audra.speech.jobs.create({
  model: "audra-essential",
  voice: "june",
  text: chapterText,            // up to 500k chars
  format: "mp3",
});

const done = await audra.speech.jobs.wait(job.id);   // polls until completed
await writeFile("chapter.mp3", await done.audio());

pip: the Python client

pip install audra

from audra import Audra

audra = Audra(api_key=os.environ["AUDRA_API_KEY"])

resp = audra.speech.create(
    model="audra-essential",
    voice="june",
    text="Chapter one. The river was low that autumn.",
    format="mp3",
    idempotency_key=str(uuid.uuid4()),   # safe retries
)
open("chapter.mp3", "wb").write(resp.audio)

Native SDK vs OpenAI SDK: when each wins

• Use the OpenAI SDK (base_url https://audratalks.com/v1) when you are migrating an existing OpenAI-TTS codebase and want zero new dependencies — /blog/openai-sdk-tts-migration has the swap. You get input/response_format naming and the audio.speech surface, not the jobs API.
• Use the native SDKs when you are building fresh: first-class jobs.wait() polling (no hand-rolled GET /v2/speech/jobs/{id} loops), typed Idempotency-Key support, and field names that match /openapi.yaml exactly — text and format, not input and response_format.
• Both SDKs share the same operational envelope: sync up to 50k chars, jobs up to 500k, 60 RPM free / 120 RPM paid per key, Bearer auth, mp3 or wav output at −16 LUFS.

Whatever you pick, keep the key in env and rotate at /dashboard/keys (rotation limited to 5/hour). The Python production checklist — retries, chunking, error classes — is at /blog/python-openai-sdk-audra-tts.