> For clean Markdown of any page, append .md to the page URL.
> For a complete documentation index, see https://developers.deepgram.com/llms.txt.
> For AI client integration (Claude Code, Cursor, etc.), connect to the MCP server at https://developers.deepgram.com/_mcp/server.

## On-Premises Release 220831

Deepgram released a new version of its on-premises solution.

### On-Premises Release 220831 Docker Hub Images

- `deepgram/onprem-api:1.64.0`
- `deepgram/onprem-engine:3.33.0`
  - Minimum required NVIDIA driver version: `>=450.80.02`


- `deepgram/onprem-license-proxy:1.2.1`
- `deepgram/onprem-billing:1.4.0`
- `deepgram/onprem-metrics-server:2.0.0`

### This Release Contains The Following Changes

- Support for improved Diarization. The new Diarization model should be placed into the `model_manager` search path and can be invoked by using the `diarize=true` parameter in the ASR request. Please visit the [**Diarization**](https://developers.deepgram.com/documentation/features/diarize/) documentation page to learn more.
  - Improved Diarization beta users should contact Deepgram support to receive the latest version of the new Diarization model, after which the beta `[[features.diarizer]]` section in `engine.toml` can be removed.


- ASR requests can invoke a specific model by passing the the model UUID instead of the model name for the `model` URI parameter.
- Streaming [**Callback**](https://developers.deepgram.com/documentation/features/callback/) requests can now receive the transcript response in the requesting WebSocket (in addition to the callback) by including the `inspect_response=true` parameter in the ASR request.
- Support for the deprecated `weights` field in `engine.toml` has been removed. Use the `path` field moving forward.
- Streaming audio which is received >1.25x real-time speed is now throttled.
- Resolves a stability issue in the License Proxy Server.