Tool Calling with the AI API in India | unoblox

Tool calling — also called function calling — lets a model decide when to invoke one of the functions you describe to it, instead of only replying in prose. It is what turns a chat model into an agent that can look up an order, query a database, or fetch live data before answering. On unoblox this works with the same request shape the OpenAI SDK already uses, against one rupee-billed endpoint, so an India-based team can build agentic features without juggling separate credentials or currencies per provider.

What tool calling actually does

You pass a list of tools (name, description, JSON-schema parameters) alongside your messages. Instead of only returning text, the model can return a structured call — which tool to run and with what arguments — and your application executes it and feeds the result back in the next turn. The model itself never runs the function; it only decides when to ask for it.

Defining a tool in the request

curl https://api.unoblox.ai/v1/chat/completions \
  -H "Authorization: Bearer ub-gw-your-key-here" \
  -H "Content-Type: application/json" \
  -d '{
    "model":"openai/gpt-5-mini",
    "messages":[{"role":"user","content":"What is the delivery status for order 4021?"}],
    "tools":[{"type":"function","function":{"name":"get_order_status","description":"Look up an order by id","parameters":{"type":"object","properties":{"order_id":{"type":"string"}},"required":["order_id"]}}}]
  }'

The response, when the model chooses to call the tool, includes the function name and arguments instead of a plain-text answer — your code runs get_order_status, then sends the result back as a follow-up message.

Picking a model for tool-heavy workflows

Tool calling is broadly supported, but reliability at following a schema varies by model tier. A reasonable spread by budget:

• Everyday agents: DeepSeek V4 Flash (₹9.07 / ₹18.14 per 1M tokens) for cheap, frequent tool checks.
• General-purpose agents: GPT-5 mini (₹25.2 / ₹201.6) or Qwen3 235B-A22B (₹9.07 / ₹55.44).
• Complex, multi-step tool chains: Claude Sonnet (₹201.6 / ₹1008) or GPT-5 (₹126 / ₹1008).
• Anything not listed here, including o3-mini and Claude Haiku, still supports tools — check /models for live ₹ pricing before you commit a workload to it.

Common tool-calling patterns

• Single lookup: one tool, one call, answer the user directly with the result.
• Multi-tool routing: give the model several tools and let it pick the right one per request — useful for a support bot that can check orders, refunds, or account status.
• Chained calls: the model calls a tool, reads the result, and decides whether a second tool call is needed before replying — common in research or data-gathering agents.
• Parallel calls: some models can request more than one tool in a single turn, which your application executes concurrently before responding.

Frequently asked questions

Do I need a different request format per model provider? No. unoblox normalises the OpenAI-style tools parameter across the catalog, so the same request body works whether the underlying model is GPT, Claude, DeepSeek, or Qwen.

What happens if the model calls a tool that doesn't exist? It won't, as long as you only reference tools you declared in the request — the model is constrained to the list you pass in tools.

Is tool calling billed differently from a normal chat completion? No — the tokens used to describe your tools and read back the results are billed the same as any other input and output tokens, in ₹, on your monthly invoice.

Can smaller, cheaper models handle tool calling reliably? Smaller models can call tools, but the more steps or the more tools you offer at once, the more a stronger model tends to help — start cheap and move up only where you see failures.

Does unoblox support parallel tool calls? Where the underlying model supports it, yes — check the individual model page under /models for its specific tool-calling behaviour.

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