AI for retail in India

Retail and D2C brands in India live and die by catalogue and customer-support text at scale — thousands of product descriptions, support replies and reviews, often needed in more than one language. unoblox gives retail teams one rupee-billed, OpenAI-compatible endpoint for GPT, Claude, DeepSeek, Qwen, Llama and more, so catalogue and support tooling can run on rupee economics from day one.

Where retail teams plug in a language model

The high-value uses are volume ones: generating first-draft product descriptions from a spec sheet, drafting customer-support replies for common queries, summarising reviews into themes a merchandising team can act on, and localising catalogue copy into regional languages. A merchandiser or support lead reviews the output — unoblox doesn't decide pricing, inventory or what a customer sees; it drafts the copy your team asks for.

Catalogue-scale economics

Generating descriptions for a large catalogue is a per-token cost problem, so the cheaper models matter more here than almost anywhere else.

Model	₹ / 1M input	₹ / 1M output	Typical use
qwen/qwen3-1.7b	0 (free)	0 (free)	Free pilot for a description generator
deepseek-ai/deepseek-v4-flash	9.07	18.14	Bulk product-description drafting
qwen/qwen3-235b-a22b-instruct-2507	9.07	55.44	Regional-language catalogue localisation
openai/gpt-5-mini	25.2	201.6	Higher-quality support chatbot replies

Live pricing for the full catalogue, including GPT-4.1, GPT-4o and the Claude line, is on /models.

The same catalogue, every regional language

Localising a catalogue for Tamil Nadu, West Bengal or Maharashtra doesn't need a separate translation vendor — route the copy through a general multilingual model such as qwen/qwen3-235b-a22b-instruct-2507 or meta-llama/llama-4-scout-17b-16e-instruct, then have a native-language reviewer check tone and product terminology before it goes live.

A sample call

from openai import OpenAI

client = OpenAI(
    base_url="https://api.unoblox.ai/v1",
    api_key="ub-gw-xxxxxxxxxxxxxxxxxxxx",
)

listing = client.chat.completions.create(
    model="deepseek-ai/deepseek-v4-flash",
    messages=[{"role": "user", "content": "Write a 60-word product description from this spec sheet."}],
)

Being clear about what it doesn't do

unoblox doesn't build a recommendation engine, manage inventory, or personalise pricing — it's an inference API. If you want personalised copy, your application passes the shopper's context into the prompt; the ranking and business logic stay yours. On data: only the free qwen/qwen3-1.7b model runs on infrastructure unoblox controls, and other models run on their original providers' infrastructure outside India.

Frequently asked questions

Can it generate thousands of product descriptions cheaply? Yes — route bulk drafting to a low-cost model such as deepseek-ai/deepseek-v4-flash and pay only for tokens actually used.

Does unoblox personalise recommendations for each shopper? No. It generates text from the context you provide; recommendation logic and ranking stay in your own application.

Can we localise our catalogue into regional Indian languages? Yes, using a general multilingual model, with a native-language reviewer checking the output before publishing.

Is customer or order data stored in India? Only for the free qwen/qwen3-1.7b model. Other models run on their original providers' infrastructure outside India.

Is there a free tier to test before scaling up? Yes, qwen/qwen3-1.7b is free to call.

How is this billed through a festive-season traffic spike? Per token actually used, in rupees, on one monthly GST invoice — there's no separate seasonal plan to negotiate.

Get started in rupees → https://unoblox.ai/sign-in