Staff paste contracts and customer data into public ChatGPT
We deploy the assistant in a closed setup on your own accounts. We use model providers with no-training terms, or a model hosted on your server when you need stricter control.
Your own ChatGPT for staff: answers from your documents, data stays inside
A closed assistant for employees. It searches your policies, contracts and knowledge base, and cites the source in every answer.
Quote and schedule after your brief
Who it's for
- Companies where people spend hours searching chats, drives and PDFs
- Sales and support teams that need fast answers on catalogs, pricing and terms
- Legal, HR and finance teams that cannot paste internal data into public ChatGPT
- Teams that want faster onboarding for new hires
Problems we solve
Finding an answer takes hours, and senior people keep answering the same questions
The assistant searches your documents (RAG: retrieve relevant passages, then answer from them) and links the source. People can see where each answer came from.
Generic chatbots make things up, so nobody trusts them
The assistant answers only from your documents. If the answer is not there, it says so. Before launch we test it on a question set written by your own team.
Not everyone should see every document
We pull permissions from Google Drive, SharePoint or Confluence. The assistant will not show anyone content they could not open themselves.
What's included
- Source audit: where documents live, what format they are in, what is outdated
- Connections to Google Drive, SharePoint, Notion, Confluence, email or CRM
- Document cleaning, chunking and semantic search setup
- Model selection based on your data rules and budget
- Web chat interface or integration with Slack, Teams or Telegram
- Role and department-based access control
- Source citations in every answer
- Query log and model cost dashboard
- Test question set and quality check before launch
- Team guide and admin training
Choose the scope
Use a package to describe what you need. We agree on the quote and schedule after reviewing your brief and requirements.
01
Pilot
One department and one knowledge source, to prove the idea
- 1 document source
- Web chat for a pilot group
- Test question set and pilot KPI
- Query log
02
Company
Several departments with different access rights
- Up to 4 data sources
- Role-based permissions
- Slack, Teams or Telegram integration
- Cost dashboard and logs
- Team training
03
Private deployment
When data cannot go to external model providers
- Open model on your server or private cloud
- Single sign-on (SSO) for employees
- Extended logging and access audit
- Documentation for your security team
How we work
Brief and requirements
You describe the task and current workflow. We clarify requirements, choose a first stage and prepare a quote.
Data audit
We map where documents live, who may see what, and what to exclude. We agree the pilot KPI together.
Pilot
We launch for one group and one source. We measure answer accuracy and search time on your own test questions.
Rollout
We connect the remaining sources and departments and set permissions. Payment is in stages.
Handover
All accounts, keys and files are already yours. We train your admin and leave documentation.
Scope and limits
- We do not promise error-free answers. The assistant can be wrong, which is why every answer links its source for checking.
- We cannot fix messy documentation without your input. If your policies are outdated, the answers will be too.
- We do not guarantee instant adoption. You need training and a named owner on your side.
FAQ
Will our documents reach OpenAI or Google?
It depends on the model we choose. For cloud models we use business terms under which your data is not used for training. If that is not enough, we host an open model on your own server. That costs more and needs GPU hardware.
What do the models themselves cost?
We prepare a quote after reviewing your brief. We list support and any third-party costs separately: subscriptions, APIs, hosting or media spend where the project needs them. Provider rates are checked before work starts.
What do we need to prepare?
Access to your document sources, a person who knows their content, and 20–30 real questions your staff ask. We use those questions to test quality.
How do you deal with the model making things up?
The assistant answers only from retrieved passages and shows them. If there is no support, it says it found no answer. Errors cannot be removed entirely, but they are easy to check.
Who owns the system?
We agree on the handover of code, materials and account access before work starts. Third-party licences and services are listed separately when needed.
How does this connect to a website and customers?
The same engine can later be added to your site as a customer chat, or to your CRM to speed up replies from managers. We suggest proving it internally first, where the risk is lower.
Start with your task
Describe your business, the outcome you need and your current workflow. We will clarify the requirements and prepare a quote after your brief.