Private GPT

How your private AI deployment works

Here is the full walkthrough from discovery through launch, including where your data lives at every step. The schedule is confirmed after the data, access, integration, and review scope is clear.

Reference architecture

What happens between a question and a sourced answer

This shows the product path at a useful level. Exact services, connections, logging, model providers, and data-handling terms are confirmed for each deployment.

Illustrative, not a customer screenshot
  1. 01Access

    Authorized user asks

    The request starts in the branded application with the user identity and permissions configured for the deployment.

  2. 02Dedicated environment

    Private GPT applies access rules

    The application determines which approved sources and workflow rules are available for that request.

  3. 03Approved sources

    Relevant context is retrieved

    The retrieval layer selects supporting excerpts from the documents and systems included in scope.

  4. 04Provider boundary

    A configured model responds

    The selected context is sent to the chosen model endpoint under the provider terms documented for the deployment.

  5. 05Review

    The user receives a sourced answer

    The application returns the response with available source references so the user can verify it before use.

Configured in the deployment

Identity, permission groups, approved sources, retrieval rules, conversation context, source display, and available audit data.

Documented at the provider boundary

The selected model endpoint, content included in a request, provider retention and training terms, and responsibility for reviewing changes.

A model provider receives the context included in the configured request. Retention and training behavior follows the selected provider, endpoint, agreement, and settings. It is not a blanket privacy claim.

The deployment phases

Phase 1

Discovery and data mapping

You tell us how your business runs, which documents and systems matter, and who should have access to what. Together we build a data map covering the approved sources, permission boundaries, and workflows in scope.

Phase 2

Environment provisioning

We provision a Microsoft Azure environment dedicated to the deployment. Networking, encryption, authentication, and role based access are configured to the approved scope, with your IT team involved where its systems or policies require it.

Phase 3

Data connection and indexing

We connect the approved sources in scope, which may include SharePoint, Google Drive, file shares, CRM records, accounting exports, and internal documents. The connection method, stored indexes, permissions, and source behavior are documented before launch.

Phase 4

Branding, training, and launch

We apply your logo and colors, put the assistant on your domain, and run onboarding around the approved workflows. Launch follows testing and your signoff on the configured experience.

What actually happens to your data

This is the part most vendors hand wave. We would rather spell it out.

Storage and retrieval are documented

Approved files and indexes are configured in the dedicated Azure deployment. The design records where originals, indexes, conversation context, and logs live for the agreed connections.

Model-provider terms are part of the design

Relevant context may be sent to a configured model endpoint to produce an answer. The selected provider, endpoint, retention and training terms, and request content are reviewed for the deployment instead of hidden behind a generic privacy promise.

Access follows configured roles

Identity and permission groups determine which approved sources a user can retrieve from. Available audit data and administrator access are also defined in the deployment scope.

Ownership and exit terms are explicit

The proposal and service agreement identify the deployment environment, data responsibilities, support boundary, and what happens to stored information if the service ends.

Want the security detail? Read the full security and data privacy overview or prepare your internal rules with the AI data handling policy workbook, see how the retrieval approach compares to RAG vs fine-tuning, review the published K3 Technology case study, or see what shapes a Private GPT proposal.

See a live deployment on your screen

Book a meeting and we will walk you through a private environment, then discuss the requirements, timeline, and pricing for your company.

Book a meeting