Service status (updated 2026-11-03): available — live captions and realtime speech-to-text in 14 languages, live translation, text translation (14 languages and Tamil), batch and fast file transcription, speech detection, speaker diarization, text-to-speech (34 voices in English, Spanish, French, German, Brazilian Portuguese and Hindi, plus voice design in all 14 languages), text tools, chat and coding, vision and document OCR, image generation and editing, sound effects, music and songs. Paused — sound-stream, voice studio routes and cloning. Details in the docs.

DEVELOPER PORTAL & QUICKSTART

Integrate Ollalink in a few lines of code.

Plain HTTPS and WebSocket endpoints with one API key. OpenAI-style requests for speech, transcription, chat and images — no client library required.

1. Get a key and authenticate

Key login.ollalink.com → Platform

Sign in at login.ollalink.com, open Platform and create an API key. It starts with sk_. Keep it on your server — never in browser or mobile code.

Header X-NH-GPU-Key

Send the key in the same header on every host. No key → 403; a wrong key → 401 (HTTP) or close 1008 (WebSocket).

X-NH-GPU-Key: sk_XXXX

2. Hosts and routes

Speech runs on the gpu-*.ollalink.com hosts; text translation, text tools and on-demand models run on ai.ollalink.com.

Host	Route	What it does
`gpu-asrhq.ollalink.com`	`POST /v1/audio/transcriptions`	Batch speech-to-text for Hindi, Hinglish and other languages (multipart `file`, optional `language`, `word_timestamps`; speech → English text). A 20 s clip in about 11–13 s.
`gpu-transcribe.ollalink.com`	`POST /v1/audio/transcriptions`	Fast file transcription in en, es, fr, de, pt, ru, it, nl; audio or video up to 100 MB; English numbers as digits. Other languages → `400 unsupported_language` (use gpu-asrhq). A 20 s clip in about 3.5 s.
`gpu-transcribe.ollalink.com`	`POST /v1/audio/segments`	Speech detection: start and end times of speech.
`gpu-transcribe.ollalink.com`	`POST /v1/audio/diarization`	Who spoke when, up to 8 speakers (`response_format` json or rttm). 60 s of audio in 1–3 s.
`gpu-asr.ollalink.com`	`wss /v1/audio/transcriptions/realtime`	Realtime speech-to-text, OpenAI-realtime-style events; 14 languages as locales (e.g. `en-US`, `ja-JP`).
`gpu-live.ollalink.com`	`wss /v1/translate/stream`	Live captions: partial and final captions in the spoken language; translate each final with `/v1/translate` for live translation. en, hi, es, fr, zh, de, ar, pt, ru, ja, ko, it, nl, vi.
`gpu-tts.ollalink.com`	`POST /v1/audio/speech`, `GET /voices`	Text-to-speech: 34 voices in en, es, fr, de, pt, hi, up to 4,000 characters (Hindi voices: 600); MP3 by default. Voice design: `"voice": "design"` plus an `instructions` description (up to 400 characters) invents a voice in any of the 14 languages, up to 1,000 characters of text. `gpu-voice.ollalink.com` serves the same voices and routes.
`ai.ollalink.com`	`POST /v1/translate`	Text translation, always on: `{"text", "target", "source"?}` → `{"translation", "numbers_verified", …}`; en, hi, es, fr, zh, de, ar, pt (Brazilian Portuguese), ru, ja, ko, it, nl, vi and ta (Tamil), any direction; `"stream": true` for live captions; up to 5,000 characters; numbers, IP addresses and placeholders kept.
`ai.ollalink.com`	`POST /v1/normalize`	Spoken numbers → digits, `language` en or hi, up to 5,000 characters.
`ai.ollalink.com`	`POST /v1/embeddings`	Unit-length vectors for 1–256 inputs; `dimensions` 1024; `mode` document or query.
`ai.ollalink.com`	`GET /v1/models`	Lists the models you can call.
`ai.ollalink.com`	`POST /v1/chat/completions`	`chat` and `coding` (always on) and vision; OpenAI-style, tools and `"stream": true` supported.
`ai.ollalink.com`	`POST /v1/classify`	Classify a text with up to 20 questions (pick one of named options; yes/no and scores experimental); always on.
`ai.ollalink.com`	`POST /v1/ocr`, `GET /v1/jobs/{id}`	Document OCR, always on: PDF or image (up to 25 MB) → markdown page by page, optional block positions; runs as a job you poll.
`ai.ollalink.com`	`POST /v1/images/generations`, `POST /v1/images/edits`	Image generation from a prompt (256–1024 px, multiples of 32, optional `seed`) and editing of an uploaded picture; PNG as base64 in JSON. The default image model is always on.
`ai.ollalink.com`	`POST /v1/audio/generations`	Sound effects or instrumental music from a prompt, 1–120 s, returned as a WAV file.
`ai.ollalink.com`	`POST /v1/music/generations`	Song with vocals from a style and tagged lyrics; a job, downloaded as M4A or FLAC (non-commercial use only).

3. Chat, coding and more

chat and coding are always on. chat thinks before answering — the answer is in message.content, the reasoning in message.reasoning, and reasoning counts towards max_tokens; coding answers directly. Replies name the model chat or coding. Document OCR and the default image model are always on; vision, the second image model, audio generation, songs and speaker diarization load on first use. The model values for image and audio generation are listed in the Models docs.

Model id	What	Route
`chat`	Everyday chat, instruction following, tool calls	`/v1/chat/completions`
`coding`	Coding tasks, bug fixes, tool calls	`/v1/chat/completions`
`vision`	Vision and quick OCR (`image_url` parts)	`/v1/chat/completions`
`ocr`	Document OCR, PDF or image to markdown (job)	`/v1/ocr`
see docs	Image generation (default and second image model) and editing	`/v1/images/generations`, `/v1/images/edits`
see docs	Sound effects and instrumental music clips	`/v1/audio/generations`
`music`	Songs with vocals from tagged lyrics (job)	`/v1/music/generations`
—	Who spoke when, up to 8 speakers	gpu-transcribe `/v1/audio/diarization`

4. Quickstarts

Text-to-speech

curl https://gpu-tts.ollalink.com/v1/audio/speech \
  -H "X-NH-GPU-Key: sk_XXXX" \
  -H "Content-Type: application/json" \
  -d '{"input": "Open port 443 on the firewall.",
       "voice": "delia", "language": "en"}' \
  -o speech.mp3

Hindi / Hinglish transcription

curl https://gpu-asrhq.ollalink.com/v1/audio/transcriptions \
  -H "X-NH-GPU-Key: sk_XXXX" \
  -F "file=@class.mp3" -F "language=hi"

Fast file transcription

curl https://gpu-transcribe.ollalink.com/v1/audio/transcriptions \
  -H "X-NH-GPU-Key: sk_XXXX" \
  -F "file=@lecture.mp4" -F "language=en"

Spoken numbers → digits

curl https://ai.ollalink.com/v1/normalize \
  -H "X-NH-GPU-Key: sk_XXXX" \
  -H "Content-Type: application/json" \
  -d '{"text": "open port four forty three", "language": "en"}'
# → "open port 443"

Chat

curl https://ai.ollalink.com/v1/chat/completions \
  -H "X-NH-GPU-Key: sk_XXXX" \
  -H "Content-Type: application/json" \
  -d '{"model": "chat",
       "messages": [{"role": "user",
                     "content": "Explain what BGP does in two sentences."}]}'

Translate a live caption

curl https://ai.ollalink.com/v1/translate \
  -H "X-NH-GPU-Key: sk_XXXX" \
  -H "Content-Type: application/json" \
  -d '{"text": "Do not open port 443 before Friday.",
       "target": "ja", "stream": true}'
# → data: {"delta": "…"} … data: {"done": true, "translation": "…"}

Embeddings

curl https://ai.ollalink.com/v1/embeddings \
  -H "X-NH-GPU-Key: sk_XXXX" \
  -H "Content-Type: application/json" \
  -d '{"input": ["How do I reset a switch?"],
       "dimensions": 256, "mode": "query"}'

Who spoke when

curl https://gpu-transcribe.ollalink.com/v1/audio/diarization \
  -H "X-NH-GPU-Key: sk_XXXX" \
  -F "file=@meeting.wav" -F "response_format=json"
# → {"duration": …, "speakers": …,
#    "segments": [{"start": …, "end": …, "speaker": …}]}

5. Limits

Every request must finish within 100 s — use "stream": true for long chat answers.
Live captions and realtime speech-to-text share capacity; when it is full the server sends busy and closes with 1013 — retry with back-off.
ai.ollalink.com: 600 requests per 60 s per key.
File transcription: audio or video up to 100 MB. Text-to-speech: up to 4,000 characters (Hindi voices: 600; designed voices: 1,000 characters plus a description of up to 400). Normalize: up to 5,000 characters. Embeddings: 1–256 inputs per request. Document OCR: up to 25 MB. Images: 256–1024 px per side. Audio generation: 1–120 s.

6. Retired and paused

The live caption lane returns captions in the speaker's language; for live translation send the finals to POST https://ai.ollalink.com/v1/translate.
Other old gateway routes on ai.ollalink.com answer 410 or 404; chat uses "model": "chat" or "coding".
Paused (503): sound-stream.ollalink.com (speech-to-speech and streaming text-to-speech) and gpu-dhvaani.ollalink.com (old Indic voice host; Hindi voices are on gpu-tts).
Paused: the voice studio routes on gpu-voice (/clone, /voice_design, /custom_voice, /speakers → 404) cloning from your own clip (ref_audio / ref_text → 400). Voice design itself works on POST /v1/audio/speech with "voice": "design".
Text-to-speech outside English, Spanish, French, German, Brazilian Portuguese and Hindi, and the earlier catalog voices (nh-… ids) → 400.