← सभी गाइड पर लौटें

Build a GPT and Claude CLI with 59API

Command-line tools are one of the simplest ways to make AI useful in daily development. A small CLI can summarize a file, explain an error, generate release notes, or compare answers from GPT and Claude without requiring a web interface. In this guide, you will build the basic architecture for a Python CLI that sends prompts to either model family through 59API.

Decide What Your CLI Should Do

Start with one focused command. For example, the tool can accept a question and an optional file, then return an answer in the terminal. Keeping the first version narrow makes it easier to test authentication, model selection, input handling, and error messages before adding advanced features.

A useful command might look like this:

• aiask "Explain this traceback" --model claude-sonnet
• aiask "Write a commit message" --file changes.diff --model gpt

Use a model alias in your own code, such as gpt, claude-sonnet, or claude-haiku, and map each alias to the exact model identifier supported by your 59API account.

Prepare the Python Project

Create a virtual environment and install the OpenAI Python SDK. 59API provides an OpenAI-compatible interface, so the same SDK pattern can be used for GPT and supported Claude models.

Set your credentials as environment variables rather than putting an API key in source control. For example, define OPENAI_API_KEY with your 59API key and set the client base URL to https://api.59api.com. Environment variables also make it easy to use separate keys for local development, testing, and production.

Your application should read the key at startup and fail with a clear message if it is missing. Never print the key in exception messages, logs, or debug output.

Send a Request Through the SDK

The core implementation only needs a client, a model, and a messages list. Construct a system instruction that describes the CLI's behavior, then append the user's question. When a file is provided, read it as text and include it with a clear label so the model can distinguish instructions from source material.

Use the SDK's chat completion method with your configured base URL. The model value should come from your alias map, while the user prompt should come from parsed command-line arguments. Read the returned assistant message and write only that content to standard output. This keeps the command composable with tools such as pipes and shell scripts.

Add Model Switching

GPT and Claude have different strengths, and model selection is a valuable CLI feature. A stronger model can handle complex refactoring or long reasoning tasks, while a smaller model is often enough for classification, short summaries, and quick explanations.

• Use a high-capability model for architecture decisions and difficult debugging.
• Use a fast, lower-cost model for repeated edits or short transformations.
• Make the default model configurable with a command-line option or environment variable.
• Display the selected model before sending a request so users can understand usage and cost.

Because 59API offers pay-as-you-go access to GPT and Claude models, you can choose the model for each task instead of paying for a large fixed subscription. Its relay exposes native official-quality models without a downgrade, while its pricing is designed to be among the cheaper options for developers. Check the current model names and rates in your account before hard-coding them.

Handle Failures and Usage Limits

Network failures, invalid model names, expired keys, and rate limits are normal possibilities. Catch API exceptions and return a short explanation with a nonzero exit status. Add a timeout, and consider one limited retry for temporary server errors. Do not retry authentication failures or malformed requests.

For production use, record request duration, selected model, and token usage when the API response provides it. Avoid storing full prompts by default because they may contain source code, credentials, or personal data. A simple maximum input length also prevents accidentally sending an entire repository.

Improve the CLI Gradually

Once the basic command works, add stdin support, JSON output, conversation history, and a --dry-run option that shows the selected model without making a request. Tests should cover missing credentials, unknown aliases, file encoding errors, and successful responses using a mocked client.

59API is a practical place to start when you want one low-cost API connection for both GPT and Claude workflows. You can sign up, create a key, set the base URL, and experiment with small requests before expanding the tool. A referral rebate may also reduce costs as your usage grows.

मूल्य · दस्तावेज़