Skip to content
NoviaHub Docs

Editors and extensions

Cursor

Use NoviaHub in Cursor through the Override OpenAI Base URL option — what it covers, the setup steps and the known limits.

Cursor is a code editor with built-in AI. It lets you bring your own API key (BYOK); the officially supported providers are OpenAI, Anthropic, Google, Azure OpenAI and AWS Bedrock. To make Cursor call NoviaHub you need the Override OpenAI Base URL option.

Prerequisites

Section titled “Prerequisites”
  • A NoviaHub API key (see API keys) and some balance (see Wallet and top-ups).
  • A Cursor account (Cursor requires signing in).
  • About Cursor plans: the docs say the free Hobby plan can use Agent, Chat and Tab with the Auto model. They don’t say whether Hobby can use custom models. The docs of ZenMux, another AI gateway, say custom models in Cursor typically require a Pro subscription.
  • A model with both the Chat and Response endpoint labels (see the model details under Models & Pricing). On 2026-09-28 these were gpt-6-sol, gpt-6-luna, gpt-6-astra, gpt-5.6-sol, gpt-5.6-luna, gpt-5.6-terra, deepseek-v4-flash, deepseek-flash, deepseek-v4-flash-vision-exp and kimi-k3. Models change; the model page is authoritative.

Install

Section titled “Install”

Download from cursor.com/download:

  • macOS 12 or later: .dmg (separate builds for Apple silicon and Intel).
  • Windows 10 or later: .exe.
  • Linux: an apt repository, an RPM repository and an AppImage; see the official installation page under References.

According to Cursor staff on the forum, 3.15.x had an input-focus bug in these settings, fixed in 3.15.20; use 3.15.20 or later.

Configure NoviaHub

Section titled “Configure NoviaHub”

  1. Open Cursor Settings, go to Models and scroll to the API Keys section at the bottom.

  2. Paste your NoviaHub key (starts with sk-) into OpenAI API Key and turn on its toggle. The official docs say to click Save after pasting; Cursor staff point out that pasting the key and turning on the toggle are two separate steps, and the override in the next step only applies while the toggle is on.

  3. Turn on Override OpenAI Base URL and enter:

    https://noviahub.com/v1

  4. Add a custom model on the Models page (ZenMux’s docs call the button “+ Add Custom Model”). Enter its model ID, for example gpt-6-sol (it must match the ID on NoviaHub’s model page exactly), and check that the toggle to the right of the model is on.

  5. Start a new chat, select that model in chat or agent and start working. Cursor staff confirmed that starting a new chat after turning the toggle on is the right move.

Verify

Section titled “Verify”

Cursor’s docs give no check button for this setup. To verify, select the model you added in chat and send a message. A normal reply means it works; then confirm the call under Usage Logs in the NoviaHub console.

Switch models

Section titled “Switch models”

Add more custom models on the Models page, then switch between them in the chat or agent model picker. Each model must meet the “both Chat Completions and Responses” condition above, and its name must not start with claude-.

To go back to one of Cursor’s own models (other than Claude and Gemini), turn off Override OpenAI Base URL and the OpenAI API Key toggle first, and turn them back on afterwards. This is the workaround Cursor staff give.

Troubleshooting

Section titled “Troubleshooting”

Tab completion doesn’t go through NoviaHub That is by design: custom API keys only apply to chat models, and Tab always uses Cursor’s built-in models.

Cursor’s own models fail with “This model does not support custom API keys” With the override on, Cursor’s own models such as Composer and Grok are sent with your OpenAI key too. Turn the override off first, as described under Switch models.

“AI Model Not Found” or “Model name is not valid” Cursor staff describe two causes. Either the OpenAI API Key toggle is off, so the request goes out without your key: delete the key and add it again, turn the toggle off and on once, then start a new chat. Or the model name starts with claude-, so Cursor used the Anthropic key instead.

Some requests succeed and some fail Check that the model has both the Chat and Response labels. Cursor sends both request formats depending on the situation, and a model that supports only one fails on some requests. ZenMux’s docs also say Cursor sometimes sends a Responses-format body to the Chat Completions address; NoviaHub then returns 400 with field messages is required. ZenMux’s docs describe this as a Cursor-side issue with no working fix yet.

GPT models fail in Agent mode with an error about tools[N].custom Cursor staff confirmed on 2026-08-10 that with an OpenAI key and the base URL override, the tool definitions in Agent requests can fail validation, with an error that mentions tools[N].custom. It is a known Cursor-side issue they are tracking.

401 Invalid token The key is wrong, or it is disabled, expired or out of quota. Check it on the API Keys page; see The four statuses.

No API Keys section on a team plan Cursor staff say that when a team admin turns on “Restrict Personal API Keys”, Cursor hides the whole API Keys section, and OpenAI has its own switch in the team settings too. Ask your team’s Cursor admin. The official docs also say that on Teams and Enterprise plans, requests with your own key still carry the Cursor Token Rate of $0.25 per million tokens.

References

Section titled “References”

Checked on 2026-09-28 and 2026-09-29: