Changelog

Five billing read endpoints are now callable with your user API key (Authorization: Bearer <key>). Previously these were UI-only; use them to build credit-balance dashboards, ledger exports, or invoice pipelines outside the console.

Base URL is https://api.smallest.ai (not /atoms/v1). Organization is resolved from the API key itself, so no X-Organization-Id header is required.

Only these read endpoints accept the API key. Purchase, auto-reload, and payment-method routes stay UI-only; a leaked key cannot make purchases or change payment methods.

See the Billing section in the API Reference for full request/response schemas.

New page under Create Agent → Build your agent that orders every agent configuration field by priority so you know what to fill in first.

Four tiers:

• Required. Name, prompt, voice, language, model. The minimum to hold a real conversation.
• Recommended. First message, phone number, knowledge base. Every production agent has these.
• Advanced. VAD, denoising, voicemail detection, timeouts, turn-taking, pronunciation. Reach for these when tuning a live agent.
• Publish and activate. The two-step gate that promotes changes to live traffic.

Complements the existing Quick start (five-minute template walkthrough) and Agent Settings overview (deep-dive card index). This page is the “what should I touch and in what order” answer for the middle ground.

smallestai==5.3.0 retires the legacy Waves STT methods and moves both file and live transcription under one client.waves.speech_to_text namespace. It also adds voice cloning, the Electron LLM completion client, and a webhook-update method. Constructor signature is unchanged: SmallestAI(api_key="...") keeps working.

Update your install

$
pip install --upgrade smallestai
$
# pinned: pip install smallestai==5.3.0

Breaking: unified speech-to-text

The two legacy STT methods are removed and replaced by one namespace. Migration is a rename, no behavior change:

Batch transcribe(...) takes typed model (pulse / pulse-pro) and language. There is no deprecation alias; the old names are gone in 5.3.0.

Typed streaming helper

The live endpoint reads its session config (model, language, sample_rate, encoding, and tuning knobs) from the WebSocket handshake query string. speech_to_text.stream(...) accepts these through request_options["additional_query_parameters"]. For a typed, discoverable call, use the helper:

1
from smallestai import SmallestAI
2
from smallestai.waves.helpers import stream_speech_to_text
3
4
client = SmallestAI()
5
with stream_speech_to_text(client, language="en", sample_rate=16000) as socket:
6
    socket.send_audio_chunk_out(pcm_bytes)
7
    socket.send_finalize_signal({"type": "finalize"})
8
    for event in socket:
9
        print(event)

The helper types the full Speech-to-Text param set: audio (sample_rate, encoding, format), transcript formatting (word_timestamps, sentence_timestamps, diarize, punctuate, capitalize, itn_normalize, numerals, full_transcript, max_words), turn-taking (finalize_on_words, eou_timeout_ms), VAD (vad, vad_events, vad_threshold, vad_min_speech_ms), redaction (redact_pii, redact_pci), and keyword boosting (keywords). Booleans are sent as "true" / "false" and list params (keywords, redact_pii, redact_pci) are comma-joined for you; anything not named still passes through additional_query_parameters.

New (additive, no action needed)

Migration

The rename is mechanical. A single repo-wide sed:

$
find . -name '*.py' -exec sed -i.bak \
>
  -e 's/\.transcribe_pulse(/.speech_to_text.transcribe(/g' \
>
  -e 's/\.pulse_stt_streaming(/.speech_to_text.stream(/g' \
>
  {} \;

If you imported smallestai.waves.types.transcribe_pulse_* directly, switch to the transcribe / stream response types. Full SDK-side write-up: MIGRATION.md on smallest-python-sdk.