One install
Applications attach. They do not replace it.
Platform
Private AI Foundation is the install. DocxIntel and AgentLab are applications on that install. You review the perimeter once. Everything that follows runs on the same audit trail.

One install
Applications attach. They do not replace it.
Open weights
Apache 2.0 and MIT only, on disk.
You hold keys
In every tier that touches your data.
No egress
Inference does not need a network path out.
Air-gapped, your cloud tenancy, or bare metal. The substrate changes. Key custody does not.
Serving, Model Gateway, governance, RBAC, audit and data protection land once.
DocxIntel reads documents. AgentLab acts on what they said, same install.
30–45 days on one document type in your environment, success criteria in writing.
Open-weight vision, LLM, embeddings and reranking, run where the data already sits.
Every prompt, completion and tool call on one controlled path. No side door.
Policy and DLP see the request before weights do. Human-in-the-loop on consequential actions.
Who may invoke which model class on which data class. Named, not implied.
Logs written where the workload runs. Disconnect the network and they still fire.
Encryption with keys you hold. Nothing is retained by BizfyLabs, because it never reaches us.
The runtime. Installed once, inside your perimeter.
Policy sees the request before weights do.
A report generated after the fact is not governance.
The substrate changes. You still hold the keys.
Applications inherit the install. You do not re-litigate the perimeter for each feature.
Apache 2.0 and MIT weights only. Licence text ships on disk. No usage gate a vendor can revise later.
We license software. We never hold production data. You are controller and processor.
Annual licence, volume-tiered applications, named fixed-fee deployment. Not per page. Not per token.
In your office, on your documents, with the cable pulled out. No competitor's sales engineer can do this.