Cursor | Vtrix API Docs

Overview

Cursor is an AI editor built around a VS Code-like development experience. It works well as an editor, but recent versions place meaningful limits on third-party API providers. Tab completion, Background Agent, Auto mode, and some built-in model routing paths may not reliably read a custom Base URL.

For Vtrix, the recommended workflow is to use Cursor as the editor and run Codex or Claude Code in Cursor’s integrated terminal. If your Cursor version still exposes OpenAI API key and Base URL override settings, you can try routing Cursor Chat through Vtrix.

Do not treat Cursor as a full agent integration

Cursor inline completion, Background Agent, and Auto mode may bypass third-party settings. This guide covers optional Chat setup and the recommended terminal workflow, but it does not promise every Cursor AI feature will use Vtrix.

Prerequisites

Cursor
A Vtrix API key, created in Settings
A Vtrix OpenAI-compatible model, such as gpt-5.5 or gpt-5.3-codex

Optional: use Vtrix Doctor

You can use Vtrix Doctor to inspect Cursor-related local settings and environment state. Cursor currently supports scan and config checks only; Doctor does not write or repair Cursor settings.

Recommended Workflow

Use Cursor as the editor

Keep using Cursor to open projects, browse code, and edit files. Do not use Cursor Tab completion or Background Agent as proof that Vtrix routing is active.

Run an agent CLI in Cursor's terminal

Open Cursor’s integrated terminal and run a Vtrix-connected CLI in your project directory:

# Use Codex
codex

# Or use Claude Code
claude

Codex and Claude Code read the current project directory, edit files, and run commands directly. Cursor provides the editor experience, while the agent CLI owns model calls and task execution.

Reuse the dedicated guides

This path is more reliable because each CLI explicitly reads its own config file and Base URL instead of depending on Cursor’s internal model routing.

Optional Cursor Chat Setup

If your Cursor version still shows OpenAI API key and Base URL override settings, you can try the following. Setting names may vary across versions.

Open Cursor model settings

Open Cursor Settings and find the Models or API Keys settings page.

Enter the Vtrix OpenAI-compatible endpoint

In the OpenAI-related settings, enter:

OpenAI API Key: YOUR_VTRIX_API_KEY
Override OpenAI Base URL: https://cloud.vtrix.ai/llm/openai/v1

If the settings page lets you add custom models, add Vtrix model IDs:

gpt-5.5
gpt-5.3-codex
qwen3.7-max

Disable Auto and select the model manually

Do not choose Auto in Chat. Select the added model manually, then send a simple message to test.

If Cursor still forces official models, cannot select the custom model, or says the current version does not support third-party providers, use the recommended workflow above.

Known Limitations

Feature	Status	Notes
Cursor Chat	Optional	Depends on whether the current version still exposes OpenAI API key and Base URL override settings
Tab completion	Not recommended	Often handled by Cursor’s official model path, not by third-party settings
Background Agent	Not recommended	Third-party provider support is limited
Auto mode	Not recommended	May bypass custom Base URL settings and fall back to official models
Doctor one-step setup	Not supported	Vtrix Doctor only scans and checks Cursor

Troubleshooting

Cursor does not show the custom model

Cause: The current Cursor version does not expose a custom model list, or Chat is still using Auto mode.

Fix: Disable Auto and select the model manually. If it still cannot be selected, use Codex or Claude Code in Cursor’s integrated terminal.

Tab completion is not routed through Vtrix

Cause: Cursor Tab completion is usually served by Cursor’s official model path. It is not the same as Chat’s OpenAI API configuration.

Fix: Do not use Tab completion to validate Vtrix routing. For stable agent behavior, use Codex or Claude Code.

Error: 401 Unauthorized

Cause: The API key is incorrect, or Cursor is not using that key for the current Chat request.

Fix: Copy the Vtrix API key again, confirm the Base URL is https://cloud.vtrix.ai/llm/openai/v1, and manually select the Vtrix model.

Doctor says Cursor cannot be repaired automatically

Cause: Cursor settings and version behavior are not stable enough for Vtrix Doctor to write them automatically.

Fix: Use agent-doctor check cursor for read-only checks. For stable setup, configure Codex or Claude Code first.