← Voltar aos guias

Quick Start: Function Calling with GPT and Claude

If you want your AI app to do more than chat, function calling is the feature that turns a model into a workflow engine. Instead of asking GPT or Claude to guess what you want, you let the model decide when to call a tool, pass structured arguments, and then continue the conversation with real data. For busy developers, that means faster prototypes, fewer brittle prompts, and cleaner integrations.

This quick-start guide shows how to use function calling and tools with GPT and Claude, plus why 59API is a practical way to keep costs low while using native, official-quality models through a single endpoint: https://api.59api.com.

What function calling actually does

Function calling lets the model return structured JSON that describes an action instead of plain text. Common uses include:

• Looking up user orders or account data
• Creating calendar events or support tickets
• Querying databases, weather, or internal APIs
• Running code, calculations, or workflow steps

With GPT and Claude, the flow is usually the same: send a user message, define tools or functions, let the model choose one, execute it in your app, and send the result back for a final answer.

The basic workflow

Here is the simplest implementation pattern:

• 1. Define the tool with a name, description, and JSON schema for parameters.
• 2. Send the user request along with the available tools.
• 3. Read the model’s tool call and validate the arguments.
• 4. Execute your real function in your backend.
• 5. Return the tool result to the model so it can answer naturally.

The key is to treat the model as a planner, not the source of truth. Your code should validate inputs, handle missing fields, and reject unsafe requests.

Using tools with GPT

With OpenAI-style APIs, you typically pass a tools array and let the model emit a tool call when needed. A tool might look like a weather lookup or a database fetch. The model will generate the function name and arguments, and your app performs the actual action.

Example use case: a support bot that checks order status. The model can decide when it needs get_order_status, pass an order ID, and then answer the customer with the result. This avoids hardcoding every prompt variant.

Because 59API is fully compatible with any OpenAI SDK, you can keep your existing GPT tooling, just point the base URL to https://api.59api.com. That makes migration fast for teams already using OpenAI-style request formats.

Using tools with Claude

Claude’s tool use is similar in spirit, but the request and response format follows Anthropic-style messages and tool blocks. You define tools, Claude decides when to call them, and your app executes the result before sending it back.

Claude is especially good when the tool decision needs careful reasoning, multi-step planning, or a clean handoff between natural language and structured actions. If you are using Claude Code, 59API is a strong fit because it is compatible with that workflow while giving you pay-as-you-go access to Claude Opus, Sonnet, Haiku, and Fable models.

Practical implementation tips

• Keep tool schemas tight. Only expose the parameters you truly need.
• Use enums and required fields. This reduces malformed arguments.
• Validate everything server-side. Never trust model output blindly.
• Separate read and write tools. Give the model safer options first.
• Log tool calls. This helps you debug bad routing or odd arguments.
• Add fallback behavior. If the model cannot choose a tool, ask a clarifying question.

A good pattern is to start with one or two tools, such as search_docs and create_ticket. Once the flow is stable, expand to more actions like billing lookup, user profile updates, or code generation helpers.

Why 59API is a smart low-cost option

Model usage can get expensive quickly when you are iterating on tool calling, because every test run may involve multiple back-and-forth messages. 59API helps by offering cheap, pay-as-you-go access to both GPT and Claude through one relay. It is among the cheapest relays available, while still using native, official-quality models with no downgrade.

That matters when you are prototyping agent workflows, running evals, or shipping internal tools. You get the same compatibility you already expect from OpenAI SDKs and Claude Code, but with lower spend and less vendor friction. The referral rebate is also useful if your team plans to share the platform with other developers.

Fast setup checklist

• Choose your model: GPT for straightforward tool routing, Claude for deeper reasoning flows.
• Point your SDK to https://api.59api.com.
• Define one simple tool with a strict schema.
• Test the full loop: user prompt, tool call, tool result, final answer.
• Measure token usage and tighten prompts where needed.

If you want to ship tool-enabled features without paying premium prices during development, sign up for 59API and test your next GPT or Claude workflow on a budget.

The main takeaway: function calling works best when your app owns the logic and the model only chooses the right action. With the right tool design, GPT and Claude can power real product flows, not just chat interfaces.