A private company ChatGPT with role-based access, citations and audit log
An internal assistant that knows your policies, manuals and contracts. It answers with a link to the source, sees only what the user’s role allows, and logs every request.
What hurts
Your staff already paste contract clauses and customer data into public ChatGPT. Security knows it, and a ban does not stop it. People need a tool that is as easy to use, not another memo.
Internal knowledge sits in SharePoint, Google Drive, an internal wiki and old email threads. Finding the current version of a procedure means knowing whom to ask. New hires spend weeks learning where things live.
Off the shelf cloud assistants do not know your access model. Finance must not see HR files, and a branch must not see head office documents. Without that, a bank or clinic will not sign off on any AI integration.
What we build
- Connectors to SharePoint, OneDrive, Google Drive, your internal wiki and file servers
- Document indexing into a Qdrant vector database with scheduled refresh
- SSO login (Microsoft Entra ID, Google Workspace) with role and department permissions
- Cited answers: document, section and a link to the original
- Audit log of questions and answers for your security team, with export
- Model on our vLLM servers or a dedicated server, no data sent to third party models
- Web interface, plus a Microsoft Teams or Telegram bot if needed
- A test set of real questions to check quality before launch
What you get
- A working assistant on your domain with SSO login
- Source code, infrastructure scripts and admin access
- Access matrix and documentation for adding new sources
- Question test set with reference answers and a run report
- Admin training and a short guide for employees
How we measure the result
- Share of answers with a correct source link on the test set
- Weekly active users and number of questions
- Share of questions where the assistant honestly said “I don’t know”
- Time to find a procedure before and after launch, by team survey
Systems and stack
Industries
Questions
How much does a private ChatGPT cost?
From $8,000. The price depends on the number of sources, document volume, how complex the permissions are and where the model runs: our servers or yours. We give a fixed estimate within 2-3 days after discovery and put it in the contract.
Will it make things up?
AI sometimes gets things wrong, so we configure it to answer only from retrieved passages and always show the source. If the documents hold no answer, it says it does not know. We check quality on a test set before launch and after every change.
Where do our documents go?
Documents are indexed on our servers or on a server dedicated to you. The model runs on our own vLLM, so data does not reach third party AI providers. We hold ISO/IEC 27001:2022, sign an NDA and a GDPR DPA before any access.
How do permissions work?
The assistant inherits permissions from your systems: Microsoft Entra ID groups, Google Workspace or folder rights. A user only gets answers from documents they could already open. Every request lands in the audit log.
How long does it take?
Usually 6-10 weeks. Discovery takes 1-2 weeks, then development with weekly demos. We launch for one department first, collect feedback and then roll out across the company.
What happens after launch?
We support the system under an SLA: refresh the index, add sources and watch answer quality. Questions the assistant failed go into the test set, so the same mistake does not come back.
Related services
More in this section
Describe the task. We reply within one business day, the first call is with the founder.
Describe the task. The first call is with the founder.
We reply within one business day. 30 minutes to tell you whether it is worth doing and what it costs.