Galon AI
Open to every account: top up the project balance, create a key with ai:inference, call a model.
An API compatible with chat completions. Keys carry the ai:inference scope. Each request is reserved on your balance before the model is called (402 if the balance does not cover it) and settled at the usage actually received. Responses never name a hosting provider.
Models
GET /v1/ai/models lists the models actually served, under their real names, with publisher, category, licence, context length and price in USD. Use the returned id as model.
Chat completions
curl "$GALON_API_BASE/ai/chat/completions" \
-H "Authorization: Bearer $GALON_API_KEY" \
-H "Idempotency-Key: chat-0001" -H "Content-Type: application/json" \
-d '{"model": "<id from /ai/models>", "messages": [{"role": "user", "content": "Hello"}]}'An OpenAI-compatible (Chat Completions) client works by pointing its base URL to your Galon API base followed by /ai and sending the Idempotency-Key header. A field that is not supported is refused with code, field and a message saying what to change, never silently ignored.
Streaming, tools and structured output
stream: true returns text/event-stream, with usage in the last chunk; an interrupted stream is billed for the usage received only. tools, tool_choice, response_format and image input are accepted only for a model whose catalogue entry publishes the capability; these options are available on request.
Caps and quotas
Default quotas and request bounds are on the Limits page. A spending cap per key can be set lower. Errors: 402 balance, 429 quota, 503 model_temporarily_unavailable (retry later).