← Voltar aos guias

Gemini CLI Through 59API: A Practical Relay Workflow

Start with the right connection, not just an API key

Using Gemini CLI through a unified relay lets you keep a familiar terminal workflow while managing access and spending through one provider. 59API offers low-cost, pay-as-you-go access to Gemini and other model families, with support for Gemini CLI, Claude Code, Codex, and OpenAI SDK clients.

The important distinction is protocol compatibility. Gemini CLI is not simply an OpenAI client with a different interface. An OpenAI-compatible endpoint alone does not establish that Gemini CLI will work; the relay must also handle the Gemini API requests your installed CLI sends.

This walkthrough uses 59API’s host, https://api.59api.com, and keeps model identifiers and any required endpoint paths tied to the provider’s current configuration instructions.

1. Prepare a small, safe test workspace

Use a supported Node.js LTS release, then install the official CLI with npm install -g @google/gemini-cli. Run gemini --version and gemini --help to confirm installation and inspect the options available in your release.

Start in a disposable directory containing a short README rather than a production repository. This makes authentication and routing failures easier to isolate, and prevents your first test from sending proprietary project material.

Create a 59API account, obtain an API key, and confirm that the account has usable balance. Check the current Gemini CLI setup instructions for the Gemini-compatible base URL and the exact API identifier of your chosen model.

2. Configure API-key authentication and relay routing

In Bash, enter the key without placing its value directly in shell history: read -r -s -p '59API key: ' GEMINI_API_KEY; export GEMINI_API_KEY. Press Enter after pasting the key. The hidden input does not remove the need to protect your terminal session.

Gemini CLI supports a custom Gemini base URL through GOOGLE_GEMINI_BASE_URL. If 59API’s current instructions specify the bare host, set export GOOGLE_GEMINI_BASE_URL='https://api.59api.com'. If they specify a Gemini-specific path, use that complete documented value instead. Do not append an OpenAI-style path by assumption.

Launch the CLI using Gemini API-key authentication rather than Google account sign-in or Vertex AI authentication. If the CLI displays an authentication selector, choose the Gemini API-key option. Existing saved authentication settings may need changing before the environment key takes effect.

Keep the key out of Git, shared screenshots, and project instructions. Environment variables are convenient, but they are not a substitute for secret-management controls on shared machines.

3. Verify one request before opening your repository

For a quick trial, Gemini 2.5 Flash is a sensible model to evaluate for short explanations and routine repository questions. Gemini 3.1 Pro is another option to compare on more demanding analysis. Confirm current prices and availability before choosing.

Use the relay’s exact API model identifier, not necessarily the human-readable display name. Replace YOUR_MODEL_ID in gemini --model YOUR_MODEL_ID --prompt 'Reply with exactly: relay connected'.

A successful answer confirms basic connectivity, not every capability. Check the 59API usage record for the request and selected model. That provides stronger routing evidence than simply receiving text in your terminal.

4. Move into a real development task

Open a clean working branch in a small repository and start with a bounded request: Read the README and package manifest. Explain how to run the tests. Do not modify files.

Then ask for one specific change, such as adding a missing validation test. Review tool permissions before allowing file changes or shell commands. Inspect the resulting diff and run tests yourself; relay access does not change the need for code review.

• Limit context: Exclude secrets, generated files, and irrelevant directories using the CLI’s supported ignore configuration.
• Limit scope: Request a plan before a large refactor.
• Measure spending: Compare usage across representative tasks, not just one short prompt.

5. Troubleshoot before scaling up

For authentication errors, check the key, balance, and selected authentication mode. For missing-model errors, verify the API identifier and account access. For 404 responses, recheck the base URL and protocol path. Tool-call failures after successful text responses require a separate compatibility check.

59API positions itself among the cheapest relays and states that it supplies native, official-quality models without downgrading. Its unified billing and referral rebate can add value, but compare total task cost and output quality yourself. If that workflow fits, sign up and validate a small Gemini CLI session before expanding usage.

Preços · Docs