Agents

Cursor

Conditional native desktop BYOK setup; separate from coding extensions.

Setting up with an AI assistant? Copy the setup prompt or open this guide as Markdown.

Setup prompt

Read https://zrelay.net/docs/ai.md and https://zrelay.net/docs/md/agents/coding-agents/cursor.md first.

Configure Cursor desktop for Zrelay using documented settings only. Detect my OS and installed Cursor version without installing anything. Read https://cursor.com/help/models-and-usage/api-keys and https://keys.zrelay.net/api/catalog. Preserve existing settings and explain the global OpenAI base-URL override. Use Cursor Settings > Models: OpenAI API Key, Override OpenAI Base URL = https://api.zrelay.net/v1, and an exact enabled catalog model matching my key's provider/pool and Cursor's supported API format. Do not invent settings.json keys, modify internal databases, rename models to bypass routing, or assume reasoning/Agent compatibility. Let me enter the secret directly in Cursor's local UI; never read, print, capture, or put it in this prompt. Automate only documented nonsecret options; otherwise give exact UI steps. Check the catalog's output limit (currently 4096); report unsupported controls instead of inventing them. Validate configuration without paid calls. Ask explicit consent before installation, Verify, or paid inference. Explain Cursor server routing and unsupported Tab/Auto/cloud features. If native compatibility cannot be established, link the separate Codex or Claude Code extension guide without changing Cursor's native engine. Report what remains untested.

Manual setup

1. Open Cursor Settings > Models > API Keys/OpenAI API Key. Preserve the previous values: Override OpenAI Base URL affects OpenAI requests globally, not only your added model.

2. Enter the key yourself in the local field. Enable the override with https://api.zrelay.net/v1. Save; treat Verify as a potentially billed action.

3. Add and select the exact enabled model in Chat/Agent instead of Auto. Catalog example: gpt-5.5, GPT/Flex, Chat Completions and Responses. Cursor documentation limits OpenAI BYOK to non-reasoning chat models; catalog availability alone does not establish Cursor compatibility. Do not invent a protocol selector or rename a model to bypass routing.

4. Review endpoint/model without capturing the secret. Disable the override to restore the previous route. For separate extensions follow the Codex or Claude Code guide; their panels do not change Cursor native inference.

Configuration and commands

Use the exact fields in the manual above in the client’s supported settings UI. This client has no file-based recipe here; preserve existing connections and enter the key locally.

Alternative: ask an AI assistant

Copy this complete prompt into your coding assistant. It describes configuration work; installation and paid tests need your authorization.

Setup prompt

Configure Cursor desktop for Zrelay using documented settings only. Detect my OS and installed Cursor version without installing anything. Read https://cursor.com/help/models-and-usage/api-keys and https://keys.zrelay.net/api/catalog. Preserve existing settings and explain the global OpenAI base-URL override. Use Cursor Settings > Models: OpenAI API Key, Override OpenAI Base URL = https://api.zrelay.net/v1, and an exact enabled catalog model matching my key's provider/pool and Cursor's supported API format. Do not invent settings.json keys, modify internal databases, rename models to bypass routing, or assume reasoning/Agent compatibility. Let me enter the secret directly in Cursor's local UI; never read, print, capture, or put it in this prompt. Automate only documented nonsecret options; otherwise give exact UI steps. Check the catalog's output limit (currently 4096); report unsupported controls instead of inventing them. Validate configuration without paid calls. Ask explicit consent before installation, Verify, or paid inference. Explain Cursor server routing and unsupported Tab/Auto/cloud features. If native compatibility cannot be established, link the separate Codex or Claude Code extension guide without changing Cursor's native engine. Report what remains untested.

Limits and verification

The gateway rejects requested output above 4096; no official Cursor output-budget control was established. This may block use. Keys and requests pass through Cursor servers. Tab, Auto, Cloud/Background Agents, Automations, CLI and Cursor API/SDK are outside this setup. Plain chat, streaming and tools were not tested.

These recipes were reviewed against documentation and public catalog data. No client installation, authenticated request, paid inference, streaming or tool loop was executed. Validate configuration first; a later short text test does not establish complete agent compatibility.

Recheck the live catalog and your key’s provider/pool. The reviewed examples have 128000 input and 4096 output maxima unless a client limitation above prevents controlling the request. Never print, log or commit credentials.