SDKs

Both SDKs are generated from the backend’s OpenAPI schema, so they cannot drift from the real routes. Every route decorated with sdk_expose is in them — care, voices, generations, organization usage and the speech surfaces.

pip install eesi-sdk

npm install @eesi/sdk

The packages are not published to PyPI/npm yet. Until they are, install them from a checkout of the repository — pip install -e sdk/python or npm install ./sdk/typescript.

Client

from eesi_sdk import EESIClient

client = EESIClient(
    base_url="https://api.eesi.ai",
    api_key="sk-eesi-...",   # or EESI_API_KEY
    timeout=30.0,
)

import { EESIClient } from "@eesi/sdk";

const client = new EESIClient({
  baseUrl: "https://api.eesi.ai",
  apiKey: process.env.EESI_API_KEY,
});

Request bodies are typed models, not dicts — the generated client calls model_dump() on whatever you pass, so a plain dict raises before the request is sent.

from eesi_sdk._generated_models import CarePatientCreateRequest

patient = client.create_patient(
    body=CarePatientCreateRequest(
        name="Margaret Chen",
        phone_number="+15551234567",
        timezone="America/Los_Angeles",
    )
)

What the generated clients cover

Surface	Examples
Care	`create_patient`, `create_schedule`, `call_now`, `list_calls`, `get_call`, `list_alerts`, `dashboard`
Voices	`list_voices`, `get_voice`, `create_voice`, `delete_voice`
Speech	`create_speech`, `create_transcription`, `list_speech_models`, `chat_completions`
History and usage	`list_generations`, `get_usage_breakdown`, `get_billing_credits`

Multipart uploads and audio responses are handled: create_speech returns the audio bytes, and create_voice and create_transcription take a file. The gaps are the streaming paths — SSE transcription, streamed chat completions, and /v1/realtime itself, which is a WebSocket and not an OpenAPI operation. For those, use an OpenAI client against the base URL or the raw socket; see Speech API and Realtime.

Errors

Everything the SDK raises subclasses EESISdkError, so you can catch the categories separately or together. ApiError is any non-2xx response and carries status_code plus the parsed body; RequestTimeoutError names which phase timed out.

from eesi_sdk.errors import ApiError, RequestTimeoutError

try:
    client.get_call("does-not-exist")
except ApiError as e:
    print(e.status_code, e.body)
except RequestTimeoutError as e:
    print(e)

GET requests retry once on a connection error; nothing else is retried, because a retried POST is a second call placed or a second voice cloned.

Regenerating

After changing routes or an sdk_expose decorator:

./scripts/generate_sdk.sh

That regenerates the OpenAPI slice, the request/response models, the client mixins and the docs spec. CI runs it and fails if the diff is non-empty, which is what keeps the SDKs honest.

Build

Cookbook

Billing

Client

What the generated clients cover

Errors

Regenerating

Going further

Authentication

Speech API

​Client

​What the generated clients cover

​Errors

​Regenerating

​Going further

Authentication

Speech API

Client

What the generated clients cover

Errors

Regenerating

Going further