Skip to main content

Workspaces and clean rooms

This is where people and agents actually work. Every other topic on this site describes a capability. This one describes the surface those capabilities are reached through, and the boundary that governs them while they are.

A workspace is a whole working environment, not a folder. It carries its own storage namespace, its own messaging inbox, its own members, its own capacity, its own residency, its own audit chain, and a console that can be projected through the lens of whoever is looking at it.

Start here​

The boundary​

PageWhat you get
The boundary modelWhat a workspace is, and what a clean room adds on top of one
Choosing a boundaryThe one decision that matters, answered by counting trust domains rather than by judging sensitivity
Default workspacesEvery entity gets a home at creation, with an inbox, a feed and a storage namespace of its own
What is in a workspaceThe complete catalog of tools in the console, grouped as it groups them, and the four layers that decide whether you see one
Clean roomsAliasing before compute, M-of-N signed notebooks, and an ordered egress gate that defaults to deny
Clean-room governanceThe seven surfaces a room adds, the privacy ledger that gates, default-deny egress, and the bind-never-re-author model

Working in one​

PageWhat you get
ExperiencesOne console projected through six experiences, with governance tabs that never switch off
StorageA dedicated namespace, S3 keys, managed tables, and cross-cell access as a request rather than a reach
Monitoring and activityThe workspace's own hash-chained ledger, its log tail as a membership right, automations, health and capacity
Capacity and placementBinding a workspace to real cells, and dedicating a whole service group to one workspace

Moving and recovering one​

PageWhat you get
Lifecycle, cloning & DRClone it copy-on-write or deep, archive it, move it between organizations, fail it over and fail it back

Pointing it outward​

PageWhat you get
ConsultingThe consulting experience, pointing the same machinery at an estate the platform does not yet run

Three kinds of workspace​

KindWho it is forHow it starts
A home workspaceOne entity: a person, an agent, a service account, a device, an organization, a nodeProvisioned automatically at entity creation. Nobody asks for it
A collaboration workspaceA team that trusts each otherCreated deliberately. Members have full access inside it, and isolation is between workspaces
A clean roomTwo or more parties who must collaborate and do not trust each otherCreated deliberately, deny-all by default, as a mutual-distrust overlay on a host workspace

The deciding question between the last two is never how sensitive is this data? It is how many trust domains are in the room? One means a workspace. Two or more that must collaborate without trusting each other means a clean room.

What a workspace carries​

Every workspace, including an automatically provisioned home, comes with all of this in place rather than as setup steps you perform afterward.

  • A dedicated object-storage namespace, so isolation is structural rather than a naming convention.
  • A messaging inbox and default streams, so the workspace and its owner are addressable the moment they exist.
  • Membership and access, by role (owner, editor, contributor, viewer) or by group grant, enforced at the gateway.
  • An access profile (open, restricted, sensitive, regulated) classifying what the workspace is for.
  • A residency selector, re-resolved on every placement and egress decision, so failover never crosses the pinned boundary.
  • A capacity binding to a service group, which can be a guaranteed reservation up to and including the whole group.
  • Its own hash-chained activity ledger, its own log tail, and its own lineage view.
  • Portability: a workspace is a recoverable, movable unit across eight planes, not a folder that happens to have data in it.

One environment, many views​

A workspace console is wide, because the platform is. Rather than showing every capability to everyone, or splitting the product into separate applications with separate bills and separate audit trails, the console projects one flat set of capability tabs through a chosen experience.

The invariant that makes that safe is gate, don't omit: the union of every experience's tabs plus the always-on governance set equals the full tab set, so organizing the console can never make a capability unreachable, and that property is held by a test rather than by convention.

Switching experience changes what you see. It never changes what governs you, and an analyst and a security engineer looking at the same workspace are looking at the same workspace. See Experiences.

Why this is the interesting boundary​

Most platforms isolate by customer. This one isolates by workspace, and the difference shows up in three places.

An agent is a member, not an integration. An agent holding the right grant provisions and operates resources at workspace and clean-room scope, through the same gate a human operator crosses. It is not calling an API from outside the boundary; it is inside it.

A clean room is a structural partition, not a filter. Collaborators see only per-room aliases, notebooks are alias-rewritten before compute so the kernel never sees source table names, code is signed by M-of-N reviewers before it can run, and every output crosses an egress gate that defaults to deny.

The boundary is portable. Clone it copy-on-write or deep, archive and restore it, move it between organizations, fail it over and fail it back. Eight planes move together, and a full deep clone consults the decision gate and can be refused, because duplicating everything is a high-consequence act.