PaletteAI Inference Launchpad Cursor Configuration
This page lists the configuration values Cursor uses to connect to a PaletteAI Inference Launchpad appliance. The Cursor tab of the console's Connect a coding agent panel lists these values. For the steps to set them, refer to Use PaletteAI Inference Launchpad with Cursor.
Settings
Set the following in Cursor under Settings > Models.
| Setting | Description | Example value |
|---|---|---|
| OpenAI API Key | The API token generated in the console. It begins with lpai_. | lpai_YOUR_TOKEN |
| Override OpenAI Base URL | The PaletteAI Inference Launchpad inference endpoint. Use the appliance address with the /v1 path appended. | https://amd.spectrocloud.com:8443/v1 |
| Model list | The model Cursor requests. Enable a GPT model, and then select it in chat. | gpt-5.6 |
| Add model | The control that adds a model name Cursor does not already list, such as a uniquely named alias an operator created on a client's Routing tab. | launchpad-glm52 |
Endpoint URL
Cursor uses the OpenAI-compatible API, which the gateway serves at /v1. Set Override OpenAI Base URL to your
appliance's address, the same host you use to reach the console, with /v1 appended. This differs from Claude Code,
which uses the Anthropic Messages API and adds the path itself.
Network Reachability
Cursor sends every model request from its own cloud servers rather than from your machine. Two consequences follow, and both are limits of Cursor's bring-your-own-key support rather than of the appliance.
- An appliance that only resolves inside a private network never receives the request. No client-side setting changes this. For the connectivity options Cursor offers, refer to Cursor's Private Connectivity documentation.
- An appliance that presents only a platform-issued certificate is not trusted by Cursor's cloud. Importing the platform CA into your operating system trust store lets Cursor on your machine trust the appliance, but it does not change what Cursor's cloud trusts.
Platform CA Certificate
When the appliance serves HTTPS with a certificate issued by its own platform Certificate Authority (CA), the connect
panel adds a CA certificate step and an instruction to import that certificate into your operating system trust
store. Cursor is a graphical application, so it reads the operating system trust store rather than a shell variable, and
it picks up the certificate only after a restart. The browser saves the file as palette-ai-inference-launchpad-ca.crt
in your Downloads folder.
The import enables Cursor on your machine to trust the appliance, so Cursor's own client-side validation of the base URL and the API key succeeds. It does not enable the model request itself. Cursor's cloud servers send that request, and they never consult your machine's trust store.
Model Name
The appliance seeds a gpt- alias prefix in every tier map, so the GPT model the panel names reaches the appliance. The
seeded prefix resolves for a client that inherits the appliance-wide table, and a client with its own Tier map can leave
it unmapped. For the rule, refer to Routing Behavior.
Cursor decides where to route a request by the model name, so a name that also exists in Cursor's own catalog can route
to Cursor's backend instead. In that case, an operator can add a uniquely named alias on the client's Routing tab,
as described in
Manage a Client's Model Access. Both
the aliases and the ids of the models the appliance serves appear in the console model list and in the appliance's
/v1/models API response.
Supported Modes
| Cursor mode | Supported | Notes |
|---|---|---|
| Ask | Yes | Routes to the appliance through the model you enable. |
| Agent | No | Locked to Cursor's own models for bring-your-own-key setups. |
| Edit | No | Locked to Cursor's own models for bring-your-own-key setups. |
| Tab | No | Locked to Cursor's own models for bring-your-own-key setups. |
Token Quotas
If the token's quota is exhausted, the appliance returns an HTTP 429 response and Cursor surfaces the error. For how
quotas apply, refer to Clients and Quotas.