A small, familiar surface.
MindLarp uses an OpenAI-compatible chat completions interface over JSON and HTTPS.
Base URL
For this local application, the base URL is http://localhost:3000/v1. Use your public HTTPS origin in production.
List models
GET
/v1/modelsRequires the models:read scope. Returns the available model catalog.
Create chat completion
POST
/v1/chat/completionsRequires the inference:write scope.
Body parameters
| Field | Type | Description |
|---|---|---|
model | string · required | A MindLarp model identifier. |
messages | array · required | Messages with role and content. |
stream | boolean | Stream the upstream response when supported. |
temperature | number | Sampling temperature, from 0 to 2. |
max_tokens | integer | Maximum output tokens, from 1 to 32768. |
{
"model": "MindLarp-Sol-1-Plus",
"messages": [
{"role": "system", "content": "Be concise."},
{"role": "user", "content": "Why is the sky blue?"}
],
"temperature": 0.7
}Inference configuration
Set MINDLARP_UPSTREAM_URL to your OpenAI-compatible model service and MINDLARP_UPSTREAM_KEY to its credential. Without an upstream, the endpoint returns a clear 503 inference_not_configured error instead of a fabricated answer.
Errors
Errors use a consistent JSON object with a message, type, and optional machine-readable code. See Errors & limits.
