What's in place, and what isn't yet
Change the mandate below: the decision changes, and the rule that applied shows up. It's a simulation, with no model behind it. Then read where the product really stands.
Try the mandate
Clone mandate
Outside the mandate: always declined
When the answer stays inside the mandate, the twin catches up with the person.
Quality lead cloneFictional example
Ask it something
Answered, sources cited
Rule applied: cite sources
Ten working days from receipt of the report, on form PQ-07-F2, approved by the quality lead.
SourcesPQ-07 §3.1Interview, March 12
The logic behind this demo
- A knowledge question: answered, and cited if the setting requires it.
- Anything leaving the organization: follows the sharing setting.
- A decision with an owner: handed to that person.
- Out of mandate: declined, whatever the settings.
This demo runs in your browser with made-up documents. No data is sent and no model is called.
Product status
Last documented review: October 7, 2026. Most of these checks need an open workspace.
| What we say | Status | How to check |
|---|---|---|
| The clone keeps the decisions you walk it through, as sources you can edit. | In place | The decision workshop and the clone's Sources tab. |
| Sensitive actions wait for a decision in an approval inbox. | In place | Approval inbox, from the Team plan. |
| Every answer always cites its sources. | Partial | The model is told to, for federated knowledge. It isn't yet a guarantee checked on each answer. |
| A request outside the clone's signed contract gets no answer: it goes to a person. | In place | Ask a question in a sphere missing from the clone's contract. |
| AI provider keys stay in the system keychain on the desktop. | In place | Desktop app settings, once a provider is connected. |
| You bring your own providers: OpenAI-compatible, Anthropic, Ollama, llama.cpp. | In place | Provider settings in your workspace. |
| Slack, Discord and Telegram gateways carry text, never your keys. | In place | Gateway configuration and the data flow register. |
| An OpenAI-compatible API and MCP servers expose your clones. | In place | Developer space, from the Business plan. |
| Certifications are signed receipts, and hosted attempts leave execution receipts. | In place | The clone's history, after a first certification. |
| A hosted clone answers third parties on the governed path (level L2 required). | Not yet | No clone is certified above L1 to date. |
| Desktop app with signed installers. | Partial | The app works; signed installers are in preparation. |
| SOC 2 compliance. | Not yet | Readiness work only, no audit report. |
| Production customers, testimonials and logos. | Not yet | None to quote. We'll show them only when they're real. |
Where your data goes
The desktop app keeps its main data on your machine, but it isn't purely local. A signed-in workspace exchanges certain data, described in the privacy notice, with our servers. Models, transcription, calendar, gateways and updates are separate flows, each following its own configuration.
Compared with the others
Knowledge capture products are built mainly for querying what's known. Population simulators predict groups. TwinRelay governs what one person's delegate is allowed to do. If you only want to query knowledge, a capture product may suit you better.