Use PaletteAI Inference Launchpad with Cursor
This guide explains how to connect Cursor to a PaletteAI Inference Launchpad appliance so that a model running on the appliance serves your requests instead of a cloud provider. You create a uniquely named model alias on the appliance, point Cursor at the appliance, and confirm that requests route through the appliance.
Cursor routes only Ask mode (chat) requests to a custom endpoint. Agent, Edit, and Tab remain locked to Cursor's own models. This is a limitation of Cursor's bring-your-own-key support, not of the appliance, so those modes do not use the appliance even after you complete this guide.
Prerequisites
- Cursor installed and already working. For installation, refer to the Cursor documentation.
- A running PaletteAI Inference Launchpad appliance with at least one model deployed and serving. To deploy a model, refer to Deploy a Model.
- An API token for the appliance. To create one, refer to Generate an API Token. Generating the token and creating the model alias later in this guide can require operator access.
- The appliance reachable at a DNS name with a valid, publicly trusted TLS certificate. Cursor sends requests from its own cloud servers, so a self-signed certificate does not work and there is no client-side workaround.
Create a Model Alias
Cursor sends model requests from its own cloud servers and decides where to route each request by the model name. If the
name matches a model already in Cursor's catalog, such as glm-5.2 or gpt-4o, Cursor routes the request to its own
backend and never contacts your appliance. To force Cursor to use the appliance, serve the model under a unique alias
name that does not exist in Cursor's catalog.
Creating an alias is an operator task. If you do not have operator access, ask an administrator to create the alias and give you its name, then continue to Configure Cursor.
The appliance can serve any model id as an alias of a model it already runs. On the appliance, alias a unique name to a
served model. Replace <appliance-host> with your appliance address and <admin-session-token> with an operator
session token.
curl --silent "https://<appliance-host>/admin/apply" \
--header "Authorization: Bearer <admin-session-token>" \
--header "Content-Type: application/json" \
--data '{"proposed_op":{"op":"set_tier","alias_prefix":"launchpad-glm52","model":"zai-org/GLM-5.2","thinking":"off","confirmed":true}}'
Set alias_prefix to the unique name Cursor requests, and set model to the id of a model the appliance already
serves, such as zai-org/GLM-5.2. To find the served model id, check the appliance's /v1/models endpoint or the admin
view in the console.
The alias then appears in the appliance's /v1/models response and routes to the real model. The alias_prefix value,
such as launchpad-glm52, is the name you enter in Cursor.
Configure Cursor
Point Cursor at the appliance in Cursor's settings. For the full list of settings and their example values, refer to Cursor Configuration.
The console lists these same steps for you. Select Connect coding agent and open the Cursor tab for the base URL and key to enter. Use the unique alias from Create a Model Alias as the model name, not a built-in model name that Cursor already knows.
-
In Cursor, open Settings > Models.
-
Enable Override OpenAI Base URL, and enter your appliance address with the
/v1path appended, such ashttps://<appliance-host>/v1. -
In the OpenAI API Key field, enter your
lpai_token. -
Under OpenAI API Key, select Add model, and enter the unique alias name, such as
launchpad-glm52. A unique name forces Cursor to treat the model as your custom model and send the request to your endpoint. -
Turn off Cursor's built-in models so that only your alias is active.
Verify the Connection
Confirm that Cursor routes a request to the appliance instead of to its own backend.
-
In Cursor, open a chat and set the mode to Ask.
-
In the model picker, select your alias, such as
launchpad-glm52. -
Send a test prompt.
reply with exactly CURSOR_OK and nothing else -
Confirm that Cursor displays the reply
CURSOR_OK.
Because the alias name exists only on your appliance, Cursor cannot serve it from its own backend. A reply confirms that the chat request reached the appliance, that a model there served it, and that the base URL, token, alias, and routing all work.
If Cursor returns We're having trouble finding the resource you requested, the alias name matches a model in Cursor's
catalog, so Cursor routed the request to its own backend and never contacted the appliance. Create the alias again with
a more unique name, and select the new name in Cursor.
Request Routing and Quotas
Requests you send through the appliance are subject to the routing rule and quota configured for your client. If an
operator configured a routing rule for your token, the appliance can redirect a request to a frontier model instead of a
local one. Quotas apply to the client, so every API token that belongs to the client draws on the same limits. When the
client exhausts a quota, the appliance returns an HTTP 429 response.
To understand what a client is and how API keys and quotas govern usage, refer to Clients and Quotas. To understand how the Tier map and the semantic router combine to pick a model for each request, refer to Routing Behavior.
Next Steps
To look up the base URL, alias, and the Cursor modes the appliance supports, refer to Cursor Configuration.