Cursor / Cline Custom API: Base URL Setup Guide

AI coding tools push their own subscriptions, but nearly all of them keep a "custom OpenAI-compatible endpoint" escape hatch. Paste the right Base URL and you can run any model per-token instead of paying a flat monthly fee. Here's the per-tool setup.

Two addresses to remember#

• Base URL (OpenAI-compatible): https://api.zaimod.com/v1
• API Key: create one in the console, starts with sk-

Tools using the Anthropic-native protocol take https://api.zaimod.com (no /v1).

Cursor#

1. Settings → Models → in the OpenAI API Key section enable Override OpenAI Base URL
2. Base URL: https://api.zaimod.com/v1, paste your key
3. Verify, then add model names manually (e.g. claude-sonnet-5, gpt-5.6-terra)
4. Switch models in Chat/Composer as needed

Cursor's Tab completion runs on their own model and is unaffected.

Cline (VS Code)#

1. Open Cline → settings gear
2. API Provider: OpenAI Compatible
3. Base URL https://api.zaimod.com/v1, your key, Model ID claude-sonnet-5
4. For native Anthropic: Provider Anthropic, Base URL https://api.zaimod.com

Cline contexts get long — claude-sonnet-5 or gemini-3-pro-preview (1M context friendly) hold up best.

Roo Code#

Same shape as Cline. Roo's trick: assign different models per mode — a cheap model (gpt-4o-mini) for Ask, a premium one for Code. That alone cuts costs by half or more.

Continue#

Edit ~/.continue/config.json:

{
  "models": [
    {
      "title": "Claude via zaimod",
      "provider": "openai",
      "model": "claude-sonnet-5",
      "apiBase": "https://api.zaimod.com/v1",
      "apiKey": "sk-your-zaimod-key"
    }
  ]
}

Cherry Studio#

1. Settings → Model Provider → Add provider
2. API address: https://api.zaimod.com/v1
3. Paste the key, add models manually (e.g. gemini-3.1-flash-lite-preview, deepseek-v4-flash)
4. Enable streaming for the typewriter effect

ChatBox#

Settings → Model → provider OpenAI compatible, API Domain https://api.zaimod.com, paste key, type the model name.

Troubleshooting checklist#

1. Missing or extra /v1: OpenAI-compatible paths need /v1; Anthropic-native doesn't
2. Model name typos: copy from the model list, don't type from memory
3. Key limits: confirm the key isn't disabled or quota-capped in the console
4. 401: wrong key, or a leading space
5. 429: rate limit — back off and retry; persistent 429s go to support

What it costs#

Heavy coding use (300 interactions/day, 3K input + 800 output tokens average):

• All claude-sonnet-5: about $4.8/day
• Q&A on gpt-4o-mini ($0.16/$0.62) + coding on claude-sonnet-5: about $1.5/day

Per-token with per-mode model assignments usually beats a $20/month subscription — and there's no quota to race for.

No key yet? Sign up and be configured in two minutes →

Further reading#

• What Is the OpenAI-Compatible API?
• Quickstart: your first API key
• Full price list (live)