The assistant we are building
on the governance layer.
We are building an assistant into the permission and audit layer VRS already runs on. Suggestion-by-construction: it will read with your permissions and propose — it cannot alter sealed pay, because the layer it runs on makes that impossible. In development with early-access customers. The conversations below are scripted demos of the intended experience – scroll and watch them answer, then press the buttons yourself.
Why this isn't ChatGPT with your logo on it.
Reads your data, not everyone's
Your chat session accesses exactly the data your role has permission to see. No more, no less. A worker sees their own hours, leave balance and upcoming shifts. A manager sees their team's data. An HR admin sees the full employee record. The boundary sits at the API level, where a clever prompt can't talk its way past it.
Remembers what matters
AI Chat maintains memory across sessions. It knows you prefer daily summaries, that you're the acting foreman on Q9, and that you asked about L. Brandt's overtime twice last week. Context and persona carry over, so you don't explain yourself every time you open the chat.
Human always in the loop
AI Chat never writes to your data without your explicit confirmation. It proposes; you confirm. Approve a leave request, send a correction, notify a worker: every action is a button you press, not something that happens behind the scenes. The audit trail records who triggered what.
Operational questions, in plain language.
For managers and operations leads
Manager · Anna BergConnectedManager · Anna BergConnectedManager · Anna BergConnectedFor workers
Chat · Thomas Hagen · workerConnectedChat · Thomas Hagen · workerConnectedChat · Thomas Hagen · workerConnectedIt knows what a rest-window violation is.
Permission-scoped data reads
Reads only what your role can see. The API enforces it, so no prompt can widen it.
Memory & persona across sessions
Remembers your role, preferences and past questions. No re-introduction every session.
Norwegian-first
Answers fluently in Norwegian and English. Mix languages freely in the same conversation.
Knowledge base
Company-specific policies, collective agreements and SOPs indexed and cited in every answer.
MCP / bring your own model
Connect GPT-4o, Claude, Gemini, Mistral or an on-premises model via MCP.
How VRS AI Chat accesses your data securely.
Every chat session operates through a permission-enforced API layer. The model never holds a database connection. It calls read-only VRS REST endpoints using the scoped token of the signed-in user. Responses are grounded in live data, not cached summaries.
Entra ID SSO or email/password
User authenticates to VRS.
Scoped access token
Chat session opens with a token derived from the user's role.
Minimal-scope retrieval
Each request retrieves only the data relevant to the question, not a bulk export.
Staged write proposals
User confirms via UI, VRS writes and creates an audit entry.
You own your history
Memory stored in VRS, not the model provider.
Any model provider
GPT-4o, Claude, Gemini, Mistral or custom endpoint via MCP.
GET /api/v1/me/timebank Authorization: Bearer <scoped-session-token> → { "flex_balance": "4:30", "overtime_balance": "12:15", "leave_remaining": 14, "period_cap_remaining": "2:45" } POST /api/v1/proposals/leave-request { "worker_id": "usr_eh_4821", "date": "2026-05-29", "type": "annual_leave", "confirmed_by_user": true } → 201 Created audit_entry_id: "ae_9f3c2a..."
The Agent doesn't wait to be asked.
It watches your operation and brings you recommendations with a confidence score and the evidence behind them. You're the manager here – review all three. Approve, adjust or dismiss: every decision is yours, and every decision is logged.
AI proposes · a human confirms · the audit trail remembers. That order never changes.
Bring your own AI, or use ours.
VRS AI Chat connects to any major model provider via the Model Context Protocol (MCP). On-premises and air-gapped deployments are supported.
Be part of the first wave.
VRS AI Chat is in active development with a small group of early-access customers. Book a demo and we'll onboard your team when your configuration is ready, usually within 2–4 weeks of the demo.
No commitment · We'll reach out within 5 business days