← Volver a las guías

Build a Dual-Model CLI with GPT and Claude

Why build a dual-model CLI?

A command-line tool powered by GPT and Claude is useful when you want the speed of automation without leaving the terminal. In practice, the best CLI assistants do more than answer prompts: they summarize logs, rewrite code snippets, generate commit messages, and explain failures from pasted stack traces. The trick is choosing the right model for the right task.

That is where a relay like 59API becomes practical. It gives you pay-as-you-go access to Claude models, including Opus, Sonnet, Haiku, and Fable, plus GPT models, all through a single OpenAI-compatible endpoint. Because it works with Claude Code, Codex, and any OpenAI SDK, you can build once and swap models without changing your whole toolchain.

Start with a simple architecture

For a real-world workflow, keep the CLI small and opinionated. A good setup has four parts: input parsing, model routing, prompt templates, and output formatting. The CLI can accept commands like summarize, rewrite, explain, and plan. Then route each command to the model that fits the job.

• GPT for quick transforms, structured output, and fast back-and-forth tasks.
• Claude Sonnet or Opus for deeper reasoning, longer context, and better analysis of large files.
• Claude Haiku for low-latency responses when you want a cheap, snappy assistant.
• Fable for lightweight content or creative generation where available in your workflow.

Because 59API is one of the cheapest relays and uses native, official-quality models without downgrade, you can experiment with this routing strategy without burning through budget too quickly.

Set up the API once

If you are using the OpenAI SDK, configuration is straightforward. Point the base URL to https://api.59api.com and set your API key in an environment variable. That means your CLI can keep using familiar SDK calls while the requests are routed through 59API.

A typical setup looks like this in principle: export your key, set the base URL, and then create a client inside your Node.js or Python app. For example, your config can live in .env as API_KEY and API_BASE_URL, with API_BASE_URL set to https://api.59api.com.

This compatibility matters if you already use Claude Code or Codex, because you do not need a separate mental model for every vendor. You can keep the same request shape, but choose the model that makes the most sense per command.

Build the first useful command

Let’s say your first command is ai summarize README.md. The CLI reads the file, trims it if needed, and sends a prompt that asks for a concise summary, setup steps, and missing docs. For this type of task, Claude Sonnet is often a strong default because it handles long text well.

Then add a second command: ai fix. This command accepts pasted logs or a file path and asks the model to identify the likely failure point, propose a fix, and explain confidence level. GPT can be a good first pass here if you want quick, structured output, while Claude can take over when the input is messy or large.

• Read from stdin for piped workflows: cat error.log | ai fix
• Support file input for local artifacts: ai explain src/app.ts
• Return plain text by default, JSON with a --json flag
• Add a --model option so users can switch between GPT and Claude manually

Add cost control and fallback logic

A CLI can become expensive if every request uses the biggest model. Keep the tool affordable by defaulting to the smallest model that still solves the task. Use Haiku or a compact GPT model for short commands, then escalate to Sonnet or Opus only when the prompt length or task complexity justifies it.

This is another reason 59API is a smart choice. Pay-as-you-go pricing is ideal for a utility CLI because many requests are small, bursty, and unpredictable. You are not locking your users into a heavy subscription workflow just to explain a shell error or rewrite a function docstring.

Also add fallback logic. If the first model times out, retry with a different model in the same family or shorten the prompt before giving up. In practice, a CLI that degrades gracefully feels much more reliable than one that only works in ideal conditions.

Test it on real workflows

Before shipping, test the CLI on actual developer tasks instead of toy examples. Run it against a real repository, a long stack trace, and a messy changelog draft. Measure response time, output quality, and token usage. You will quickly see which commands need Claude’s depth and which ones are faster and cheaper with GPT.

That testing loop is where a relay with official-quality models pays off. You get realistic behavior during development, which makes your prompt design, routing rules, and user experience much easier to trust in production.

Ship, then refine

Once the core commands work, add niceties like shell completion, streaming output, history, and config profiles for different teams. If your CLI is used by contractors or small dev teams, the referral rebate from 59API can also help offset usage as adoption grows.

If you want to build a CLI that feels fast, modern, and affordable, sign up for 59API and wire it into your first command today. You will keep the OpenAI-compatible workflow you already know, while gaining access to Claude and GPT models from a single low-cost relay.

Precios · Docs