MiniCrow modelsThink, hear, speak, watch and talk. One API key.

Five models for real-world AI products, built in India — with one key, one prepaid balance in rupees, and the cost of every call in its own response.

Try them freeRead the API docs

Free in the Playground today · Public launch 1 October 2026

An AI product needs more than one sense.Here, they all share one key.

  • One API key
  • One prepaid balance
  • Prices in ₹ and $
  • The cost in every response
  • OpenAI-compatible

Explore the lineup. Use one, or all five.

  • The Osprey mascot.Live

    Osprey

    Thinks · LLM for real-world applications

    Human-like conversation, tool calling and vision — for assistants, companions and agents.

    From ₹6.9per million input tokens

    Learn moreChat with it
  • The Lark mascot.Live

    Lark

    Hears · speech-to-text API for 21 languages

    Speech to text in 21 languages — phone calls at 8 kHz, code-mixed speech, a file or live.

    From ₹20per hour of audio

    Learn moreTranscribe yours
  • The Pica mascot.Live

    Pica

    Speaks · text-to-speech API for Hindi and English

    Natural Hindi and English voices that stream as they speak, billed per second of audio.

    From ₹21per hour of audio

    Learn moreHear the voices
  • The Lark-V mascot.Live

    Lark-V

    Watches · video summary and timeline API

    One video in; a summary, a timeline of its moments and every word said, out.

    Per clipbilled on what each clip used

    Learn moreTry a clip
  • The Osprey Live mascot.Live

    Osprey Live

    Talks · real-time multimodal AI model

    Hears, sees, thinks and talks back in real time — on a phone call, on glasses, in a robot.

    From ₹65.74per call-hour, all in

    Learn moreHear a call

Also on the same key: Embed and Rerank

Dense and sparse vectors from one call, and a reranker for a shortlist — search and RAG over your own documents. minicrow-embed · minicrow-rerank

₹2.1 per Mtok · dense + sparse

Get the highlights. Each model, at work in real life.

  • Live video call

    How do I look for the interview? Be honest.

    Honestly? Sharp. The navy blazer works — just straighten the left side of your collar. And smile; you've got this.

    Osprey
  • Contact centres

    Every customer call transcribed as it happens — for summaries, quality checks and agent assist.

    Lark
  • AI companions

    A companion that talks with you in a warm, human voice — and laughs at your jokes.

    Pica
  • Creators & YouTube

    A long video, summarised with a timeline of its moments — ready for chapters and search.

    Lark-V
  • AI glasses

    What am I looking at?

    This is India Gate, a war memorial in New Delhi.

    Osprey Live
  • AI calling

    A phone agent for support, sales and appointments — it hears the caller, calls your tools and answers out loud.

    Osprey Live

Illustrative scenes. The tools are your own functions, called by the model.

Hear them. Watch them. Real audio, real outputs.

Pick a model and press play: a phone agent taking a call, a recording and its transcript, a video read end to end, a voice, a chat — in your language.

Hear it in your language

Osprey Live

Choose a scenario

≈₹66 / call-hour, all inYour tools, mid-callHindi · English
Call in progress

Live call · hears, thinks and speaks on one WebSocket

Illustrative call

Spoken by Pica: Tanvi is the agent, Aarav the caller. The conversation is illustrative.

नमस्ते! क्लिनिक से बोल रही हूँ। आपको किस दिन का अपॉइंटमेंट चाहिए?

कल शाम पाँच बजे के बाद मिल जाएगा?

check_appointment_slots · tomorrow, after 17:00

कल साढ़े पाँच बजे का स्लॉट खाली है। बुक कर दूँ?

हाँ, कर दीजिए।

send_sms_confirmation

हो गया — कन्फ़र्मेशन SMS भेज दिया है।

Read the voice agent API

Try them yourself. No account, no card.

Every voice, your own recording, your own video, a conversation — on the real models, free, right here.

Prefer more room? Open the full Playground

You build the harness. We run the senses.

Your product, your prompts, your tools and your memory are yours to write. Hearing, thinking, speaking and seeing run behind one API — so a voice app is three calls, not three vendors.

You buildyour code

The harness

  • Your productThe app your users open — on the web, on a phone, on a device.
  • Prompts and personalityHow it thinks, what it says, how it sounds.
  • Tools and workflowsYour APIs and the actions it is allowed to take.
  • Memory and knowledgeWhat it remembers, and the sources it answers from.
  • EvalsHow you know it works, call after call.

# one turn of a voice app

audio → Lark → text

text + prompt + tools → Osprey → reply

reply → Pica → voice

We runapi.minicrow.com

The senses

One keyOne billPrepaid in rupeesCost in every response

Which one is right for you? Start from what you are building.

One key. Every model, the same way.

  • One base URL and one key for chat, speech, voice, video and search.
  • The OpenAI SDK works for chat, transcription and speech — change the base URL and the key.
  • Prepaid in rupees, no subscription. A key at zero is refused before any model is called.
  • Every response says what it cost, in the same call.
Read the docsEvery price
Four models, one keybash
# one key, one base URL — every model
export MINICROW_KEY=mc_YOUR_KEY  BASE=https://api.minicrow.com/v1

curl $BASE/chat/completions \
  -H "Authorization: Bearer $MINICROW_KEY" -H "Content-Type: application/json" \
  -d '{"model":"osprey-flash","messages":[{"role":"user","content":"Aaj ka plan kya hai?"}]}'

curl $BASE/audio/transcriptions \
  -H "Authorization: Bearer $MINICROW_KEY" -F file=@call.wav -F model=lark-mini

curl $BASE/audio/speech -o hello.wav \
  -H "Authorization: Bearer $MINICROW_KEY" -H "Content-Type: application/json" \
  -d '{"model":"pica-nano","input":"नमस्ते, आज मौसम बहुत अच्छा है।","voice":"david"}'

curl $BASE/video/summaries \
  -H "Authorization: Bearer $MINICROW_KEY" -F file=@clip.mp4 -F model=lark-v-large

Questions.

Which MiniCrow model should I use?

Osprey for chat, assistants and agents; Lark to turn speech into text in 21 languages; Pica to turn text into a natural voice; Lark-V to summarise a video; Osprey Live for a real-time voice agent on a phone call or a device; Embed and Rerank for search over your own documents. Most voice products use three together: Lark hears, Osprey decides, Pica answers.

Do all the models use the same API key?

Yes. One key and one base URL, https://api.minicrow.com/v1, for every model, with one prepaid balance in rupees. There is no subscription, and every response carries its own cost.

Are MiniCrow's models OpenAI-compatible?

Osprey chat, Lark transcription and Pica speech use the OpenAI request shapes: point the OpenAI SDK at https://api.minicrow.com/v1 with a MiniCrow key. Lark-V has its own endpoint for a video in and a summary out, and Osprey Live is one WebSocket for a whole voice agent.

How much do the models cost?

Osprey from ₹6.9 per million input tokens, Lark from ₹20 per hour of audio, Pica from ₹21 per hour of audio, Osprey Live ₹65.74 per call-hour, all in. Prices are prepaid in rupees, shown in dollars too, and the pricing page lists every rate live.

Can I try the models for free?

Yes. The Playground runs the voices, transcription, video summaries and chat free in the browser, without an account — on this page and at minicrow.com/playground. API keys open to everyone on 1 October 2026.

Which languages do they support?

Lark hears 21 languages, including code-mixed Hinglish and phone calls at 8 kHz. Pica and Osprey Live speak Hindi and English today, with more Indian languages coming.

Ready to cut your voice AI bill?

Build on MiniCrow and pay up to 86% less than Sarvam.

A prepaid key in rupees, one OpenAI-compatible base URL, and a response that tells you which lane answered, why, and exactly what it cost. No plans to pick and no postpaid invoice.

Start building free

Or read the API docs first.