Cursor

Quick

Cursor is configured through a screen, not a file: Settings → Models → API Keys. Three fields, all on one screen.

# Cursor → Settings → Models → API Keys
api_key  = "$KUMO_API_KEY"     # OpenAI API Key
base_url = "https://api.kumorouter.com/v1"    # Override OpenAI Base URL
model    = "<model>"         # Add custom model

1. 1Turn on your own OpenAI keyPaste your Kumo key into the key field.
2. 2Override the base URLEnable "Override OpenAI Base URL" and enter `https://api.kumorouter.com/v1`.
3. 3Add a modelPut an identifier from the catalog into the add-model field.
4. 4Press VerifyCursor calls the gateway itself and tells you whether the key was accepted.

Different Cursor releases name the screen differently — Models or Providers. Look for the pair "OpenAI API Key" and "Override OpenAI Base URL": that pair is the connection point.

Models

Cursor does not know the Kumo catalog and will not show it for you. Models are added by hand: one identifier per line, a canonical name or an alias from the catalog.

Cursor's built-in models do not go anywhere in the meantime. Switch off the ones that should not reach the editor's own backend — otherwise the model picked in the chat decides where a request goes, and you will not notice.

Verify

Check the gateway before touching the editor: then a failing Verify button means a Cursor setting, not the key. You paste the key into Cursor's field itself, while the command below reads it from a shell variable — so export the key once in the terminal you verify from.

export KUMO_API_KEY="kumo_sk_..."

curl https://api.kumorouter.com/v1/chat/completions \
  -H "Authorization: Bearer $KUMO_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{
    "model": "<model>",
    "max_tokens": 64,
    "messages": [{ "role": "user", "content": "reply with OK" }]
  }'

Then ask one question in Cursor's chat with the added model selected, and find that call in the console log.

When it does not work

• Verify fails — the key or the address. Repeat the curl above first: if it answers, the problem is in Cursor's fields.
• 404 — the base URL is missing the /v1, or carries an extra path. The address ends exactly at /v1.
• Model not found — the identifier is not from the catalog, or was never added to Cursor's model list.
• The request is not in the log — the chat has a built-in Cursor model selected, not the one you added.

How a model is named → Every status the gateway returns →