Know where
the work goes.

One gives you choices about compute and providers. Understanding each boundary is the starting point for a private working setup.

Your workspace and compute

The product uses an API, worker, database, and configured bot computers. Persistent workspace information and files live in the services and storage chosen for that deployment. Business is proposed with semi-private compute; Enterprise is proposed with private compute. The actual isolation and operating arrangement are agreed during setup.

Your model and speech providers

When you use a hosted model or speech service, requests may include the context, audio, or text needed for that task. Those requests go to the provider you configure. A private computer does not make an external model endpoint local.

Your calls and connections

Phone carriers process call traffic. Speech and language services process the configured parts of a conversation. Connected applications receive the requests and data needed for their tools. Review access, recording, and retention choices for your environment.

Your keys and access

The product stores supported provider credentials in an encrypted secret store. Browser authentication, workspace authorization, bot tool access, and private engine controls are separate boundaries. Configure each deliberately and keep secrets out of instructions, public code, and logs.

Providers on the hosted service

Which providers handle a request depends on what you connect. On the hosted service we operate, the current arrangement is: bot computers run as local containers on our servers; managed app connections go through an integration platform (Composio) that holds the connection's tokens; model, speech, and image requests go to the provider you connect with your own key or sign-in, and no deployment-wide model key is configured; phone calling is not enabled for live calls; and website analytics is not configured. The Privacy Policy lists every provider category and what each receives, and this section is updated whenever the hosted arrangement changes.

A precise promise

“Data never leaves” is only accurate when every relevant model, speech, compute, storage, and tool path is contained in an agreed environment. The default architecture permits external providers, so we do not make that blanket claim.

For a constrained deployment, bring your requirements to Lughlabs. We’ll map the paths, supported providers, and gaps before agreeing a solution.

Read the Privacy Policy, theTerms of Service, and thethird-party licenses for the commitments behind this page.