Cursor with EXAIV: Custom API Endpoint and Model Setup Guide
Configure EXAIV as a custom OpenAI-compatible endpoint in Cursor: the Override Base URL field, API key entry, adding model IDs, verification steps, and the debugging order for 404/401 errors. Run Cursor's AI chat on domestic open models.
Cursor with EXAIV: Custom API Endpoint and Model Setup Guide
Cursor is the most popular AI code editor today, and beyond its built-in models it lets you point its "OpenAI" channel at any compatible endpoint — which means Cursor's AI chat can run on models served through EXAIV (Kimi, GLM, and more domestic open models), with one key to manage. This guide gives field-level setup steps and a verification method.
Setup steps
Get an API key
Generate an API key in the EXAIV console (shown once in full — save it immediately).Open Cursor model settings
Cursor Settings → Models page, find the OpenAI API Key section.Override the base URL
Enable the Override OpenAI Base URL toggle and enterhttps://exaiv.com/api/ai/v1.Enter the API key
Paste the EXAIV key into the OpenAI API Key field.Add a model
Click Add model, enter the model ID you want (for examplekimi-k3), save, then select it in the chat model picker.
Model IDs follow the catalog
Available model IDs follow EXAIV's live /models catalog. The model name you enter in Cursor must be a real ID from the catalog — pasting a HuggingFace weight name (like moonshotai/Kimi-K3) will fail at request time.
Verify connectivity
After setup, start an ordinary chat in Cursor (do not edit code first) to confirm the model responds. If it fails, debug in this order:
| Symptom | Check | Fix |
|---|---|---|
| 401 | Key copied fully, no stray spaces | Regenerate and paste again |
| 404 | Base URL is exactly /api/ai/v1 (nothing more, nothing less) | Correct the URL path |
| Model not found | Is the model ID in the catalog | Verify the real ID on /models |
| Timeout | Network path, oversized request | Shorten context and retry |
Scope of the custom endpoint
Some Cursor capabilities (such as Tab completion) are tightly bound to official models; a custom base URL mainly applies to chat-style requests. Check Cursor's current docs — do not assume every feature can route through a custom endpoint.
Recommended model pairings
| Use | Model | Notes |
|---|---|---|
| Code Q&A, error explanation | kimi-k3 | Long context suits pasting large code blocks |
| Tool-style planning, multi-step tasks | glm-5.3-flash | Steady tool calling |
| Everyday completion-style chat | A lightweight catalog model | Follows the live /models list |
One key covers every catalog model; switching is just the model name — the most convenient part of a custom endpoint setup.
FAQ
Will Cursor upload my code after this setup?
Why does Tab completion still use official models?
Can I configure multiple models to switch between?
For 404 and 401, which comes first?
Where is pricing?
Related reading
- Kimi K3 and GLM Integration Guide — the full API integration tutorial
- Qwen Local Coding Workflow — local coding assistant setups
- Qwen API Troubleshooting — error debugging in detail
- Continue with EXAIV — another open-source coding assistant
- Pricing — live pricing and plans