Getting started

Service status (updated 2026-11-03)

Available: translation (14 languages and Tamil) and text tools, file speech-to-text, speech detection, speaker diarization, realtime speech-to-text and live captions in 14 languages, text-to-speech in English, Spanish, French, German, Brazilian Portuguese and Hindi plus voice design in the 14 languages, chat and coding, vision and document OCR, image generation and editing, audio generation and music. The dashboard Playground has Transcribe, Text-to-speech, Translate and Live speech controls; Live speech can translate finalized captions with the separate text translation API. Voice cloning and sound-stream are paused.

The platform has two services. Which one you use depends on the capability — each capability page tells you the host to use. One GPU API key opens both.

Service	Base host	Auth
Translation, text tools and models — text translation (POST /v1/translate), spoken numbers to digits (POST /v1/normalize), text embeddings (POST /v1/embeddings), classification (POST /v1/classify), chat, coding and vision (/v1/chat/completions), document OCR, image and audio generation, music (Models)	https://ai.ollalink.com	X-NH-GPU-Key: <key> (your GPU API key)
Speech GPU — speech-to-text (batch, file, realtime), live captions, text-to-speech	gpu-*.ollalink.com (hosts below)	X-NH-GPU-Key: <key>

Voice cloning and the realtime speech-to-speech service (sound-stream) are not available; live captions arrive in the speaker’s language and are translated with a separate POST /v1/translate call — see Models & languages. Full key detail and error codes: Authentication.

1. Get a key

You want	How to get it
GPU API key (sk_…) — the text tools on ai.ollalink.com and every gpu-* host	Self-serve: sign up at login.ollalink.com → Platform → GPU API keys → create a key. Your workspace starts as pending and a new key only works once we approve it.
Legacy shared Speech GPU key	Works until 3 November 2026; replace it with your GPU API key.

Keep every key server-side. Never ship one to a browser, mobile app, or public repo.

2. Your first text-tools request

curl https://ai.ollalink.com/v1/normalize \
  -H "X-NH-GPU-Key: $YOUR_GPU_KEY" -H "Content-Type: application/json" \
  -d '{"text": "open port four forty three", "language": "en"}'

The reply’s normalized field is "open port 443". Details: Text tools.

3. Your first Speech GPU request

Speech GPU hosts

Host	What it does	Routes
gpu-tts.ollalink.com	Text-to-speech: 48 voices in English (including Indian English tutor voices), Spanish, French, German, Brazilian Portuguese, Arabic, Japanese and Hindi	POST /v1/audio/speech, POST /tts, GET /voices
gpu-voice.ollalink.com	The same voices and routes as gpu-tts (the voice studio routes are paused)	POST /v1/audio/speech, GET /voices
gpu-asrhq.ollalink.com	Batch speech-to-text for Hindi and the languages file transcription does not cover; speech → English	POST /v1/audio/transcriptions
gpu-transcribe.ollalink.com	Fast file transcription of recordings and videos (en, es, fr, de, pt, ru, it, nl), text or subtitles; speech detection	POST /v1/audio/transcriptions, POST /v1/audio/segments
gpu-asr.ollalink.com	Realtime speech-to-text in the 14 languages	wss://…/v1/audio/transcriptions/realtime
gpu-live.ollalink.com	Live captions in the 14 languages (captions in the speaker’s language; translate them with /v1/translate)	wss://…/v1/translate/stream

Use the ollalink.com names; older networkershome.com names still resolve for a while but are being retired.

Text-to-speech

The Speech GPU lanes take the key as a header, not a Bearer token. This one returns a WAV file:

curl -X POST https://gpu-tts.ollalink.com/v1/audio/speech \
  -H "X-NH-GPU-Key: sk_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX" \
  -H "Content-Type: application/json" \
  -d '{"input":"BGP peers exchange routes over TCP port 179.","voice":"delia","language":"en","response_format":"wav"}' \
  --output speech.wav

• Send voice and language on every request. Languages: en, es, fr, de, pt. Defaults: delia for en, noa for es, noemie for fr, nele for de, manuela for pt, layla for ar, yui for ja; other languages return 400. See Voices by language.
• Format: /v1/audio/speech returns mp3 unless you ask for another response_format (wav, pcm, opus, aac, flac).
• Text length: up to 4,000 characters per request; longer text gets a 400 that states the limit.

from openai import OpenAI

gpu = OpenAI(base_url="https://gpu-tts.ollalink.com/v1",
             api_key="unused",  # the X-NH-GPU-Key header is the credential
             default_headers={"X-NH-GPU-Key": "sk_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"})

audio = gpu.audio.speech.create(model="ollalink-tts", voice="ryan",
                                input="BGP peers exchange routes over TCP port 179.",
                                response_format="wav",
                                extra_body={"language": "en"})
audio.write_to_file("speech.wav")

Speech-to-text

Transcribe the file you just made — gpu-transcribe takes a multipart upload:

curl https://gpu-transcribe.ollalink.com/v1/audio/transcriptions \
  -H "X-NH-GPU-Key: sk_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX" \
  -F "file=@speech.wav" \
  -F "language=en"

language is one of en (default), es, fr, de, pt, ru; the reply carries text, and response_format=verbose_json adds segments and word times (Speech-to-text). For Hindi and other languages use gpu-asrhq.

Paths differ per GPU lane

Only gpu-tts (POST /v1/audio/speech) is OpenAI-shaped; the other hosts use their own routes (table above). No GPU host serves /v1/models, so an SDK that probes it will get a 404 — call the documented route directly. Each capability page states the exact path.

When something fails

What you see	What it means
400	Bad request — an unsupported language or voice, a missing field, or text over the limit; the message says which
401	Unknown or revoked key
403	Missing X-NH-GPU-Key, a key not yet approved, or a key without GPU access (forbidden)
404	Wrong route for that host — each Speech GPU host has its own (hosts)
410	A route switched off on this host
429	Rate limited (600 requests / 60 s per key) — back off and retry, honouring Retry-After
503	Busy, restarting or paused — retry shortly with back-off (sound-stream is paused)

Every failure on these hosts is a clean HTTP status; fix auth on the first 401/403 instead of retrying. Every failure mode, including the silent ones: Errors & troubleshooting.

Health check

To prove a key on a speech host, list the voices — 200 with the voice list means the key and the path are good; 401 or 403 means the key is wrong:

curl https://gpu-tts.ollalink.com/voices -H "X-NH-GPU-Key: sk_XXXXXXXXXXXXXXXXXXXXXXXXXXXXXXXX"

Where to next

• Authentication — which key goes where, error codes, rate limits
• Translation and text tools — text translation, spoken numbers to digits, embeddings, classification
• Speech-to-text — batch, file transcription, realtime
• Live captions — captions of a live speaker over a WebSocket, and live translation
• Text-to-speech — 48 voices in 8 languages
• Models — chat, coding, vision, document OCR, image and audio generation, music
• Voices by language — every voice, its languages, and the default per language
• Hindi with English numbers and technical terms — Hindi transcription and numbers
• Models & languages — the full per-lane coverage matrix
• Errors & troubleshooting — status codes, WebSocket failures, and the silent ones
• OpenAPI 3.1 spec — import into Postman or Insomnia, or generate a client

Use the dashboard

Open Playground after workspace approval. Speech-to-text defaults to the English/Hindi accuracy option; the fast and legacy options remain selectable. Translate calls the text translation API. Live speech can stream a microphone or audio file, display original captions and translate finalized captions. The browser receives only a short-lived, GPU-scoped session token for speech sockets; durable partner keys stay server-side.