Skip to main content
Choose Custom image in the launch wizard when managed Hugging Face / artifact loading is not enough. You bring a container that already serves HTTP. SynapsAI schedules GPUs and proxies traffic to it.

​
What you configure

Pricing is hourly only. Per-token is not available for custom images.

​
How you call it

Managed models use /v1/.... Custom containers are reached without the /v1 prefix:
GET, POST, PUT, PATCH, DELETE, OPTIONS, and HEAD are proxied. Authenticate with the same API key as other inference. Reserved first path segments (v1, webhooks, metrics, docs, …) are not custom-container routes. If the container streams, send Accept: text/event-stream or a JSON body with stream / stream_format.

​
After launch

The model appears on Models like any other deployment: Usage, Logs, Instances, Configuration, Manage. See Manage models and Autoscaling.

Managed launch instead

Hugging Face or artifact, OpenAI-compatible /v1 endpoints.