Governed Workspace
AI that builds controlled tools, not just answers.
Describe a report in chat. You get a tested, versioned tool that runs on your own device, makes the same report every time, and stamps it so anyone can check it.
Invitation only during the pilot. For quality, compliance and finance professionals.
Talk to it like ChatGPT, Claude or Copilot. Get a tool, not advice.
Before
A monthly complaint trend rebuilt by hand in a workbook. Formulas only one person understands. No record of what produced the numbers.
After
A tool you described in chat, tested and versioned. It makes the same report every time, with a stamp anyone can verify.
- The AI writes the tool. The workspace tests and runs it, with no AI in the run.
- Your data stays on your device. The AI never sees your real report results.
- Every tool inherits the same controls from the workspace. Nothing in a chat can switch them off.
From a plain-language request to a report you can prove
- 01
Describe
Attach a sample. Say what you want to process and what report you need.
- 02
Build
The AI writes the tool in a closed, declarative language. The workspace validates it.
- 03
Try
See the result on your sample, ask for changes in chat, and save tests that must keep passing.
- 04
Activate
You say when it is ready. The version's fingerprint is recorded; a bad tool fails its tests first.
- 05
Run
Pick your real files. The tool runs on your device and produces XLSX, CSV or PDF with a stamp.
- 06
Verify
Anyone can check a report at a public page, or re-run it from the same inputs.
Built-in controls, enforced by the workspace
Tests before activation
Saved tests must pass before a tool can be used.
Version history
Every change is a new version; an earlier one can be restored.
Local audit trail
A hash-chained, tamper-evident record of changes on your device.
Notarised log
Fingerprints of versions and runs are recorded outside your device, with daily timestamp anchoring.
Grounded reports
Every number comes from supplied data or a constant whose source you confirmed.
Stamp and verification
Each report carries a run id, a code and a QR that a public page checks.
Datasets that stay put
Kept datasets are checked for changes and damage, with changes recorded.
Evidence
Files attached to a report are stored by hash and never overwritten.
Your data stays on your device
The answer to "uploading company data to an unknown tool is shadow IT": we hold none of it.
On your device
- Tool scripts, versions and tests
- Your input files and datasets
- Reports and evidence
- Conversations
- A local audit trail, optionally encrypted
On our servers
- Identity and sign-in
- Version and run fingerprints (hashes)
- The notarised log
- The public verification page
No inputs, outputs, datasets or conversations.
What the AI receives
- Your messages
- The structure of a sample
- A few rows only if you choose to share them
Sent to the model provider under your own key. Never your real report results.
Anyone can check a report
The stamp
Every report carries a run id, a short code and a QR. The public page hashes a dropped file in your browser and uploads nothing.
Re-run
Given the script, inputs, parameters and date, an inspector can re-run it and confirm the report is exactly what they produce. A run made outside the app cannot be registered, so it shows as unregistered.
What people build
Complaint and CAPA reporting
Monthly trends, registers with status and age, open and overdue lists, evidence on each record.
Calibration and training due
Due reports that keep their instrument or people list between runs.
Finance workings
Reconciliations, ageing and close status from your own files, including scanned statements read on your device.
Approval stays where it is today: send the report to your approver. The stamp lets them check what they were sent.
What we do and do not claim
Designed to support, not certified
We describe the controls and the requirements they are designed to support. We make no compliance claims.
Validation stays with you
The first release is for reporting and tracking, not GMP-critical decisions.
No AI at run time
A saved tool runs with no model call. The AI never answers questions from your real results.
Approval happens outside
The workspace proves what a report is. Your approver confirms it is right.
Join the pilot
Free for a small number of early customers. Tell us what you report on today and we will send an invitation.