What your firm has already delivered, ready for the age of AI
What an engineering office delivered over the past years usually sits in a network folder: the report nobody reopens, the sheet only its author understands, the survey buried in a subfolder. In an archive project that material becomes a calculation report, a sheet and a file that can be searched, and it starts answering in search, in the assistant and on the site kiosk, with the source one click away.
Example: three files from a 2019 project go into an archive project. The .docx report becomes a calculation report, appended at the end as a draft; the .xlsx sheet comes in with its arithmetic redone on the server; and the borehole PDF becomes a project file, with text per page and its tables. At the end, the project enters the search index and starts answering in the assistant and on the site kiosk.
What changes once the archive is in
- 01
An archive project keeps, it does not coordinate
It keeps calculation reports, sheets, models, files and the site kiosk. Stages, activities, schedule and hours stay out, because that work has already happened.
- 02
The arithmetic is redone here
Input values come from the file, and results are computed again on the server instead of being copied from whatever Excel saved last time.
- 03
Whatever is left behind is stated
Images and equations in a .docx do not survive the import, and the screen lists what was left out instead of letting it look complete.
- 04
The project enters the search, not just its files
Name, type of service and description become indexed text, so asking whether the firm has done something similar finds the project, not merely a stray document.
A dormant file on one side, an answered question on the other
What is already written goes in as the format it is in. On the other side, the same material serves office search, the conversation about the firm's projects, and the kiosk on site.
- .docx, .md, .txtbecomes a calculation report, as a draft
- filled .xlsxbecomes a sheet, with the arithmetic redone here
- .pdftext per page, with the tables read
- 3D modelelements and properties of the project
- Search with citationthe excerpt comes along, and the chip opens the project it came from
- Archive conversationask about the projects you can reach, never about the whole licence
- Kiosk on sitethe archive answers on site as well
The difference is not the model, it is what reaches it
A model answers with what it is handed
A language model does not hold your archive. It gets things right when the right excerpt arrives with the question, and that is decided by retrieval, not by the size of the model.
Three searches, one list
The same question becomes a search by meaning, one by word and one by the categories marked in the text. The three are merged by rank, because their scores are not comparable.
The reach is the person's own
Archive search looks inside the projects that person can open on screen, never the whole licence, and the conversation they start belongs to them.
The archive stays yours
Preparing an archive for AI is only worth it if the material stays under your control: where it is kept, who can read it, and which model gets access to it when a question is asked.
Files stay where you point them
On the installation's own disk, or in S3 compatible storage, such as the MinIO that ships with the product.
Reading is isolated in the database
The search tables carry a rule inside Postgres itself: every query declares licence and projects, and what was not declared does not come back.
Your archive is never training data
Nothing here trains or tunes a model on your material. It is retrieved when the question is asked, and the model reads only the excerpt search handed over.
The model is the installation's choice
Anthropic, OpenAI, a compatible service, or an open model running on your own infrastructure, with no third party key.
The model choice, and what each option implies, is on Assistant.
Start with a project that is already finished
Pick one you have delivered and see what it starts answering. We reply by e-mail.
Request access