Running models on-premises
Your data doesn't have to leave your organization. But should the model run inside, too?
One of the first questions asked when bringing AI into enterprise work is where the data goes. When a model is called, the question and its relevant context are sent to wherever the model runs. If the model sits with a cloud provider, that data briefly leaves the organization as well. For some organizations this is acceptable; for others it is not.
Where the data lives and where the model runs are not the same thing
Two separate questions need to be pulled apart here. The first is where business data is stored: records, files, memory. The second is where the model runs.
When Hexense is installed on your own server, business data and files stay on your own network. The models that generate answers today, however, belong to external providers; during a model call, the relevant context leaves your network. By connecting your own provider keys, you can run these calls under your own contract. Running the model itself inside the building is a separate decision, with its own pros and cons.
When does an on-premises model make sense?
- Privacy and regulation: In areas where data must never leave the organization, such as work involving personal data or trade secrets, the model call has to stay inside as well.
- Predictable cost: For heavy, repetitive workloads, a fixed infrastructure cost can be more predictable than paying per call.
- Latency and connectivity: On sites where network connectivity is limited or response time is critical, a local model can speed up the work.
What does it cost?
An on-premises model is not a free option. It requires hardware, installation and operation. Models that can run locally do not match the strongest cloud models at every task; the gap can become noticeable in work that requires complex reasoning. Keeping models up to date is a separate responsibility, too.
That is why the question is rarely "all or nothing." The decision should be made according to which task runs on which model.
Deciding on a policy, not a single model
Hexense's model gateway first classifies every request, then routes it to the appropriate model based on cost, speed and capability. This approach becomes even more meaningful alongside an on-premises model: work involving sensitive data stays inside, while work that is less sensitive but demands more power can go to external models the organization permits. The rule depends on organizational policy, not on the user's preference in the moment.
G-Box
For this need, Hexense is preparing G-Box: a piece of hardware that runs local models and sits on the company network. With G-Box, the model itself will also stay inside the building; it will be configured and delivered according to the organization's workload and privacy level. We share its current status in the "Where we are today" section of the platform page.
How do you make the decision?
Three questions are enough to start: Which types of data must never leave the organization? How frequent and how complex is the work done with that data? For which tasks do you genuinely need the capability of the strongest model? The answers to these questions are the first draft of the policy that determines which work is done inside and which outside.
We rewrote business applications from the ground up for AI. For your organization, team, community or startup, the AI transformation starts in a single workspace, with an entirely new experience.
A shared brain, agents that work on top of it, and capabilities that come with applications.
People, work, documents and decisions in a single context, independent of any one person.
People, organizations, projects and places with their relationships, in one context graph.
Nothing becomes permanent without your approval.
Answers rest on document evidence, with the source shown.
An action carries the context of the work; a task carries the concrete job.
AI members working on the team brain, within the permissions they are given.
Models from leading providers through a single gateway, or with your own keys.
Each request is weighed first, then sent to the right model.
The agent works out from the team brain whose question it is, which work and which decision it belongs to.
Chat, email, schedules, webhooks and connected channels.
It works without waiting for a prompt, and notices what didn’t happen.
Business rules live in the application; every operation is validated and recorded.
The agent brings the calendar, the task, the table and the chart to you the moment you need them.
You work directly on the components; the work moves forward on a single surface.
ERP, CRM and other systems stay in place as systems of record.
An agent cannot exceed your permissions; sensitive steps come to you for approval, and every operation is recorded.
Depending on the situation, the agent suggests, starts, asks for approval, executes or waits.
Deployment and data boundaries follow your organization’s policy.
The workspace is set up, sources are connected, agents are configured.
We came to team intelligence not from the outside, but from inside ERP projects.
Thirty years, five hundred companies, two patents; and the team writing this platform today.
Consulting, training and custom development; all of it, or just what you need.
From current-state analysis to agent design, step by step.
Training on working with AI, for managers and teams.
For needs the ready-made applications don’t cover, on the platform.
Tap a company to see its details on the side.
The companies we bring Hexense to market with; you can join too.
Bring Hexense to your own customers and industry.
The same platform works with each structure’s own context, rules and permissions.
ERP, email and document sources meet in a single context.
Eight areas, from planning to after-sales, in one context.
Six areas, from marketing to process management, in one context.
Each department in its own workspace; management sees governed outputs in one place.
Follow-ups, reminders and reporting are handed over to agents.
Decisions and ongoing work live in the shared context, not with individuals.
As the team grows, its organizational memory grows with it.
Every application you add brings a new capability to the shared intelligence.
A platform-independent, flexible, AI-powered ERP enhancement package.
Learning experiences with LXStudio, in the team’s context.
From running the center to AI-assisted project proposals, in a single context.
With HexRTLS, location and event data flow into the team intelligence.
Human resources management in the team’s shared context.
AI-powered customer relationship management, right in the workspace.
How Hexense gets work done: project management built into the platform.
Dynamic processes with workflow design, on the platform.
A platform fee per workspace and a fee per user.
What we write and talk about on team intelligence and enterprise AI.
The latest post is on the left; tap one to open it on the side.
The latest videos; tap one to play it on the side.
Announcements, short notes and posts on team intelligence.
Tell us briefly about your team and systems; we’ll get back to you.