Retool AI with unoblox (India)

Retool apps usually reach an AI model through a REST API resource rather than a dedicated AI block, which makes unoblox a natural fit: add it once as a workspace resource, and every internal tool your team builds afterwards can call GPT, Claude, DeepSeek or Qwen against a single rupee-billed account instead of scattered personal API keys.

Add unoblox as a Retool resource

Under Resources > Create New > REST API, set the base URL to https://api.unoblox.ai/v1 and add a header Authorization: Bearer ub-gw-xxxxxxxxxxxxxxxxxxxxxxxxxxxx — store the key using Retool's resource-level environment variables or secrets rather than typing it into an individual app.

Calling it from a query

Create a new query on that resource with method POST and path /chat/completions. The request body needs a model field (for example qwen/qwen3-235b-a22b-instruct-2507) and a messages array built from your app's text input or table selection. Bind the response's choices[0].message.content to a text component, and the internal tool is functional.

Choosing a model inside an internal tool

Model id	₹ / 1M input	₹ / 1M output
qwen/qwen3-235b-a22b-instruct-2507	₹9.07	₹55.44
openai/gpt-5-mini	₹25.20	₹201.60
deepseek-ai/deepseek-v4-flash	₹9.07	₹18.14
qwen/qwen3-1.7b (free)	₹0.00	₹0.00

Claude Sonnet (anthropic/claude-sonnet-4-5, -4-6, -5) runs ₹201.60 / ₹1,008.00 per 1M tokens (input/output) if a tool needs a stronger model — check /models for anything not listed here before assuming a rate.

One resource, many apps

Because the resource is defined once at the workspace level, individual app builders never see or paste a raw key — usage from every internal tool that queries it rolls up under one unoblox account, which makes the monthly invoice a single line item for finance rather than a reconciliation exercise across several personal accounts.

Keeping usage within limits

unoblox applies rate limits per key rather than per app, so if several internal tools share the one Retool resource, a spike in one app's usage counts against the same limit as the others. For teams running many high-traffic internal tools off unoblox, minting a second key for a heavier workload and adding it as a second Retool resource keeps one noisy app from throttling the rest — track both against the same account dashboard so finance still sees a single combined view of spend.

Frequently asked questions

Does Retool have a dedicated unoblox connector? No — a generic REST API resource with unoblox's base URL and key is the entire integration; there is no marketplace connector to install.

How do I keep non-engineers from seeing the raw key? Store it as a resource-level secret or environment variable; queries reference the resource, not the credential itself, so app editors never need to see the value.

Can different apps in the same workspace use different models? Yes — model is a field in the request body, so each query or app can set its own without touching the shared resource configuration.

Is there a free model to prototype with before committing budget? Yes — qwen/qwen3-1.7b is priced at ₹0, which is enough to wire up and test a query before pointing it at a paid model.

Does this support streaming responses inside a Retool app? Retool queries are typically request/response; for a token-by-token feel, poll or use a custom component — the resource itself doesn't need to change either way.

How does billing work across many internal tools? One unoblox account, one monthly GST invoice in rupees, regardless of how many Retool apps or keys call it, with input tax credit claimable.

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