sociopup

AI for every team, on company terms.

A studio concept for one company AI workspace: approved models, shared workflows and sensible limits. Employees get real freedom for everyday work, without moving company data into personal accounts.

Illustration of team requests passing through a company gateway to a company-hosted model, a contracted cloud model or a blocked route
Sociopup studio concept, drawn for this page. Not a client project: the company, people and data are fictional.

The problem.

Banning AI rarely stops people using it. It moves the work into personal accounts the company can’t see, support or switch off when someone leaves.

The alternative is a workspace the company runs. Staff sign in with their work account, choose from approved models and reuse prompts that already work. The rules follow the data, not a policy PDF nobody reads.

Freedom, with
a few clear rules.

Choose a team, a task and the kind of information involved. See which model is allowed, the rule behind it and what gets recorded.

Team

Task

Information involved

Simulated with a fictional company, Harbourline Group, and prepared rules. No model, gateway or account is connected, and the budgets are sample numbers.

Allowed route

Contracted cloud model

Internal material may use the cloud model covered by the company’s agreement, which should set how long prompts are kept and rule out training on them.

Workflow
Team template: first draft in the brand voice
Recorded
Who used it, which model, the cost and the data type.
Marketing budget this month
$410 of $600 left

How it would
be built.

The proposed architecture. Nothing here is running yet; a real build starts with the company’s own systems, contracts and security team.

Architecture diagram: an employee signs in to the workspace, which sends requests through a model gateway with team keys, approved models, data rules, budgets and usage records, to a company-hosted model, a contracted cloud model or a blocked route

  1. Sign in with the work account

    Single sign-on through the company’s identity provider. When someone leaves, their access goes with their account.

  2. One gateway in front of every model

    A model gateway gives each team its own key, checks which models that team may use, applies a monthly budget and records usage.

  3. Rules follow the data

    People choose the kind of information they’re working with, and the rule picks the model route. Automatic detection can help, but the choice stays visible to the person.

  4. Model choice without lock-in

    Open-weight models on company-managed servers for sensitive work; contracted cloud models where the agreement allows. Providers can change behind the gateway without retraining staff.

  5. Shared workflows

    Prompts and small workflows that work are saved to a team library, each with an owner who keeps it current.

Where it can
go wrong.

  • Self-hosting a model keeps prompts on company-managed servers, but it brings running costs, capacity planning and security updates the company must own.
  • A contracted cloud model is governed by the provider’s terms for the company’s plan. Those terms, the data region and retention settings need checking before anything sensitive is allowed.
  • Budgets and rate limits stop runaway costs, and code helpers here can suggest changes but can’t deploy them.

Getting people
to use it.

Adoption is planned like a product launch, one team at a time.

  1. Pilot

    Two or three teams with clear everyday tasks, a short list of approved models and a few shared templates.

  2. Expand

    Team champions publish workflows that work. Blocked requests are reviewed monthly, because they show where the rules or tools fall short.

  3. Scale

    New teams join within the same rules. A regular review covers who governs it, the risks mapped, what is measured and how issues are managed.

Research behind it.

External references. These are other people’s projects and publications, linked for reading; they are not Sociopup work or endorsements.

Make it yours.

This is a studio concept with a simulated demo, not a delivered project or a product for sale. If your company wants something like it, we start with a discovery of your systems, data rules and contracts, with your IT and security people involved.

Talk about private AI