Run the whole loop inside your own network.
The same stack, deployed to your own infrastructure. Speech, inference and rendering stay inside your perimeter, and no conversation data leaves the network.
Helm chart, Docker Compose
Any OpenAI-compatible endpoint
None required at runtime
File, syslog or OTLP
Air-Gapped Operation
No outbound calls are required at runtime, and model weights are mounted from your own registry.
Your Own Models
Any OpenAI-compatible endpoint, plus a local speech-to-text and text-to-speech provider. Swap providers per organization without redeploying.
Deploys As Containers
A Helm chart and a plain Compose file are both supported. A GPU node is only required for rendering and local inference.
Audit Trail
Every utterance, tool call and credential grant is recorded to your own sink, whether that is a file, syslog or OTLP.
Have a question about this surface?
We read everything you send us.