Skip to main content
Version: v1.0.x

PaletteAI Inference Launchpad Claude Code Configuration

This page lists the configuration values Claude Code uses to connect to a PaletteAI Inference Launchpad appliance. For the steps to set them, refer to Use PaletteAI Inference Launchpad with Claude Code.

Environment Variables​

VariableDescriptionExample value
ANTHROPIC_BASE_URLThe PaletteAI Inference Launchpad inference endpoint. Use the appliance address with no path. Claude Code appends /v1/messages.https://amd.spectrocloud.com:8443
ANTHROPIC_AUTH_TOKENThe API token generated in the console. It begins with lpai_. ANTHROPIC_API_KEY is also accepted.lpai_YOUR_TOKEN
ANTHROPIC_MODELOptional. The Claude alias Claude Code requests. The appliance maps the alias to the model it serves, so you do not pick the backend model.claude-opus-4-8
ANTHROPIC_DEFAULT_OPUS_MODELOptional. The alias Claude Code requests for its Opus-tier work. The appliance maps the alias to the model it serves.claude-opus-4-8
ANTHROPIC_DEFAULT_SONNET_MODELOptional. The alias Claude Code requests for its Sonnet-tier work.claude-sonnet-4-5
ANTHROPIC_DEFAULT_HAIKU_MODELOptional. The alias Claude Code requests for its Haiku-tier, background work.claude-haiku-4-5
CLAUDE_CODE_EFFORT_LEVELOptional. Sets Claude Code's reasoning effort. One of low, medium, high, or max.max

Endpoint URL​

The endpoint is your appliance's address, the same host you use to reach the console, with no path appended. The gateway serves the Anthropic Messages API at /v1/messages, and Claude Code adds that path itself, so ANTHROPIC_BASE_URL must not include /v1. If you do not know the address, ask the administrator who set up the appliance.

Model Name​

Claude Code requests a Claude alias (claude-opus-4-8, claude-sonnet-4-5, or claude-haiku-4-5), and the appliance maps that alias to the model it serves. Set ANTHROPIC_MODEL to one of these aliases. You do not select a backend model directly. Both the aliases the appliance accepts and the ids of the models it serves, such as glm-5.2, appear in the console's model list and in the appliance's /v1/models API response.

Token Quotas​

If the token's quota is exhausted, the appliance returns an HTTP 429 response and Claude Code surfaces the error.

Resources​