Spring AI with unoblox (India)

Spring AI already speaks the OpenAI wire protocol, so the fastest way to run a Java or Kotlin service against GPT, Claude, DeepSeek, Qwen, Llama, Gemma or Kimi in India is to leave the framework alone and change where it points. unoblox is a single OpenAI-compatible endpoint billed in rupees on a monthly GST invoice, so OpenAiChatModel keeps working exactly as documented — only the base URL and key move.

Point Spring AI at unoblox

In application.yml, swap the OpenAI section for unoblox's base URL and a ub-gw-… key:

spring:
  ai:
    openai:
      base-url: https://api.unoblox.ai/v1
      api-key: ub-gw-xxxxxxxxxxxxxxxxxxxxxxxxxxxx
      chat:
        options:
          model: deepseek-ai/deepseek-v4-flash

No custom RestClient, no forked starter — ChatClient.create(chatModel).prompt("...").call().content() runs unchanged. Autowire ChatModel the way every Spring AI tutorial shows; the difference is invisible to your controllers and services.

Choosing a model from the catalog

Spring AI lets you override model per call via ChatOptions, so a single service can route cheap classification traffic to one model and higher-stakes generation to another:

Model id	₹ / 1M input	₹ / 1M output
deepseek-ai/deepseek-v4-flash	₹9.07	₹18.14
qwen/qwen3-235b-a22b-instruct-2507	₹9.07	₹55.44
openai/gpt-4.1	₹201.60	₹806.40
qwen/qwen3-1.7b (free)	₹0.00	₹0.00

Claude Sonnet (available as anthropic/claude-sonnet-4-5, -4-6 and -5) is priced at ₹201.60 / ₹1,008.00 per 1M tokens (input/output). For any model not shown here, check the live rate on its page at /models before you wire it into a ChatOptions builder rather than guessing.

Streaming, tool calling and JSON mode

Because unoblox mirrors OpenAI's chat-completions contract, Spring AI's reactive stream() path, its @Tool-annotated function calling, and ResponseFormat.JSON all work without extra adapters — the framework never has to know it isn't talking to a US-hosted host. This matters for teams who built agent flows against Spring AI's tool-calling abstraction and don't want to rewrite them just to change billing geography.

Embeddings from the same account

unoblox also exposes a native /v1/embeddings endpoint, so OpenAiEmbeddingModel can point at the same base-url and key for retrieval pipelines. Check /models for the embedding entry currently listed before pinning a model id in configuration.

What actually changes for a finance team

Nothing about the request/response shape changes — what changes is the invoice. unoblox bills in rupees on a monthly GST invoice from an Indian entity, so the spend is input-tax-credit claimable and doesn't need an international card on file. That's the whole pitch: keep Spring AI's abstractions, change the accounting.

Frequently asked questions

Do I need a special Spring AI starter for unoblox? No. The standard spring-ai-openai-spring-boot-starter works as-is; you're only changing base-url and api-key.

Can different services in the same app use different models? Yes — set model in ChatOptions per call, or run separate ChatModel beans with different defaults.

Does Spring AI's streaming work over unoblox? Yes, the SSE stream format matches what Spring AI's reactive client expects.

Is there a free model to develop against before spending? Yes — qwen/qwen3-1.7b is priced at ₹0, useful for wiring up a Spring AI integration before switching to a paid model for production traffic.

How is usage billed? In rupees, on a monthly GST invoice from an Indian entity, with input tax credit claimable — no international card required.

What if a model I want isn't in unoblox's price list? Don't hardcode a guessed rate — check the live ₹ price on that model's /models page first.

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