Security & control
Expertise travels.Your data does not.
Deployable expertise is only worth something if its circulation stays controlled, by default and by human validation.
- ✓Revocable mission accessThe invitation carries a visible expiry until it is accepted; once the mission is active, access can be cut at any time, from either side.
- ✓Controlled separation of contextsWithin the client workspace, the companion can use the knowledge accumulated in its original workspace, except for rules kept internal. In contrast, work and memories produced during the mission stay in the client workspace: they do not flow back into the companion’s original workspace.
- ✓Confined mission memoriesWhat the companion learns on a mission stays in the mission until something is promoted.
- ✓Selective promotionA reusable piece of know-how can be proposed; the client’s data never is.
- ✓Human validationAn administrator of the originating space reviews every request, and can refuse it.
- ✓Collaboration oversightWho worked with whom, on what, and what was escalated to a human.
- ✓Your models, your providersClaude, Codex, Ollama or fully local — the choice is made per companion.
- ✓Local execution, inference at the providerThe app and your workspaces run on your machine. Inference runs at the provider of the model you chose — it stays inside your infrastructure only if you host that model.