Your model. One API call away.
Publish a base model, trained adapter, or supported Hugging Face model under a name your application can call. Serve it on demand or keep it ready for steady traffic.
Give the published model a stable name, such as support.
Interactive illustration · sample data
From a published model to an answer. Follow a request through an illustrative on-demand model lifecycle.
Serving that fits the way your app is used.
Choose the trade-off between idle compute cost and readiness, with a familiar interface for your application.
Use the SDK you already know
Call the OpenAI-compatible chat completions API with your Tensorant base URL, API key, and model name. Receive a complete answer or stream the response.
Explore the featureScale to zero between requests
On-demand serving uses RunPod Serverless workers. You pay while workers run, including warm time; the first request after a quiet period may wait for a cold start.
Explore the featureStay ready for steady traffic
Keep a GPU running with always-on serving for the period you choose. Extend the period when needed, and see the estimated daily compute cost before you start.
Explore the featureControl access to your models
Create application API keys in the console, restrict which published models they can call, and review usage. Try the model in the Playground before integrating it.
Explore the featureTest it. Name it. Put it to work.
Managed inference runs in your connected RunPod account and uses a RunPod network volume. Publish directly from the console, or test a model in a temporary deployment first.
- Stable model names
- Your application calls the published name through /v1/chat/completions.
- Streaming built in
- Start displaying the answer as the model generates it.
- A place to experiment
- Compare responses and generation settings in the Playground.
Build with Tensorant