Workstation on site
The dedicated workstation sits on the organisation’s LAN. We install the platform and the open-weight LLMs validated by noze ® remotely, and they run locally.
IntelliPA
From the workstation on site to signature in 5 steps, with the Reviewer before every signature.
The 5 steps01 02 03 04 05
Connected sources
We connect shared folders (SMB/Windows, NFS), IMAP mail, WebDAV, S3-compatible storage and REST/JSON sources read-only. The designated contact consents to every source and credentials are encrypted.
Offices configured
We activate the licensed offices, each with its own templates, review rules and laws. Then we open the Console to the office’s staff.
Drafts and questions
Staff prepare drafts and query the regulations. The Reviewer reads every draft again before signature.
The RUP signs
IntelliPA prepares and verifies. The RUP (the officer responsible for the procedure) or the manager decides, signs and takes responsibility for the act.
Under the hood
Local AI grounded in the organisation’s documents
Everything runs on the workstation on site: the open-weight LLMs, the RAG engine that retrieves the relevant passages and the IntelliPA web Console, reachable from the LAN with nothing to install on staff PCs. The standard workstation handles up to 8 AI requests at once and queues the others. For larger organisations it is sized to the case. IntelliPA runs on Admina (the open-source governance framework) and follows the OISG paradigm.
How we keep control
Guarantees built into the product
Reviewer before signature
It flags internal inconsistencies, outdated legal references, contradictions and formal errors, cell by cell in tables too.
Corrected document
Once the corrections are accepted, the document downloads with Word tracked changes or clean. It can also be saved to an external archive.
Answers with the source
Every answer and every draft cites the organisation’s documents it comes from. At review you verify, you do not rebuild.
Source permissions
Whoever cannot read a file at the source does not see it in search, citations, versions or exports either. An unmapped permission grants nothing.
Agents that propose
They run on a schedule or on an event, such as a new document or an approaching deadline. A person approves every change. Each automation has minimal permissions, daily limits and an emergency stop.
Signed packs and plugins
Additional content packs and plugins are signed and run in a sandbox.
Where IntelliPA sits
Compared to other approaches
| Approach | What it does | Where the data sits | Limit |
|---|---|---|---|
| Public cloud AI chatbot | Generates generic text | In the vendor’s cloud | Documents leave the organisation, no knowledge base on the acts, no consistency check |
| Copy and adapt a previous act | Manual reuse of a previous act as a template | Inside the organisation (manual) | Slow, propagates errors and outdated references, depends on the officer’s experience |
| External consultant or firm | Specialist drafting per project | With the external consultant | Costly, not always available, does not scale to daily volume |
| Cloud SaaS AI for the public sector | Vertical AI assistant | In the vendor’s cloud | Data leaves the organisation, depends on connectivity, recurring per-user fee |
| IntelliPA | Generative AI on site: generates, verifies, cites sources | On-premise, inside the organisation | Requires the dedicated workstation on site |