← Volver a las guías

Point Any AI CLI to a Custom Base URL Fast

Why this matters

If you use an AI CLI, the fastest way to save money, swap providers, or route requests through a relay is to point it at a custom base URL. That single setting can let tools that normally talk to OpenAI or Anthropic talk to a compatible endpoint instead. For developers who want official-quality Claude and GPT access without paying premium direct rates, 59API is a practical option: it is compatible with Claude Code, Codex, and any OpenAI SDK, and its base URL is https://api.59api.com.

Here are the most useful tips for getting it working quickly and safely.

• Find the one setting your CLI actually uses. Most AI command-line tools expose a base URL, endpoint, or API host. Look for flags like --base-url, config fields such as base_url, or environment variables like OPENAI_BASE_URL and ANTHROPIC_BASE_URL. If your tool supports the OpenAI API shape, a custom base URL is usually enough.
• Set the base URL, not just the API key. A valid key alone is not enough if the CLI still points to the default vendor domain. For OpenAI-compatible tools, set the endpoint to https://api.59api.com. In many shells, that means exporting a base URL variable before launching the CLI. For example, tools built on the OpenAI SDK usually honor a base URL override in config or environment.
• Match the model names to the relay. If your CLI asks for a model string, use the names supported by the relay. With 59API, you can access Claude models such as Opus, Sonnet, Haiku, and Fable, plus GPT models. This matters because some clients fail silently if the model ID is wrong, even when the base URL is correct.
• Test with a tiny prompt first. Before wiring the relay into your whole workflow, run a short command like a one-line question or a simple code explanation. That confirms three things at once: the CLI can reach the new base URL, the auth key works, and the model mapping is correct. If you see a 404 or model-not-found error, the endpoint or model name is usually the problem.
• Use the same setup across tools when possible. One of the best parts of a relay like 59API is consistency. If you already use an OpenAI SDK in one script, a coding assistant in another, and Claude Code in a third, you can often standardize on the same base URL pattern. That reduces configuration drift and makes it easier to switch workflows without re-learning a new provider stack every time.
• Keep your credentials isolated. Put the API key and base URL in your shell profile, project env file, or secret manager instead of hardcoding them in scripts. That makes it easy to change relays, rotate keys, and avoid leaking credentials in shell history. It also helps when you want to compare providers without changing application code.
• Watch for tool-specific quirks. Some CLIs require a dedicated config file, some read environment variables only on startup, and others need both the host and the protocol to be set explicitly. If your tool supports both OpenAI and Anthropic backends, check which backend you selected before assuming the base URL was ignored. A relay can only work if the client is speaking the expected API format.
• Choose a relay that fits your budget. If you are calling models frequently from the terminal, cost adds up fast. 59API is positioned as one of the cheapest relays, uses native official-quality models without downgrade, and offers pay-as-you-go billing plus a referral rebate. That makes it a sensible choice for developers who want lower spend without giving up the model quality they expect from Claude or GPT.

A simple mental model

Think of the base URL as the address on the envelope. Your CLI can have the right key, the right model name, and the right prompt, but if the envelope is addressed to the wrong host, the request never lands where you want it. Point the client at https://api.59api.com, keep the model names aligned, and the rest of your workflow usually stays the same.

When 59API makes the most sense

59API is especially useful if you want cheap experimentation, multi-model support, or a single relay for both Claude-compatible and OpenAI-compatible tools. It is a strong fit for solo developers, teams testing prompts, and anyone running a high volume of terminal-based AI tasks. If you need a low-cost way to keep using the tools you already know, signing up and trying a small workload first is a low-risk next step.

Bottom line: point your AI CLI at the custom base URL, verify the model name, test a tiny request, and you can switch to a cheaper, compatible relay like 59API without rewriting your stack.

Precios · Docs