Memory, Brain Store, and subscriptions

Every room has its own memory. On top of that come inherited memory from the parent room and subscribed expertise from other teams. The Brain Store makes Corp and Public expertise discoverable, each with clear authority: Binding or Advisory.

Memory and rooms

A room is a secured workspace with a data class, rules, and members. Every room has a memory. It holds the room’s knowledge as entries (decisions, rules, and documents), from which backed answers are built.

A room’s knowledge scope (its «file cabinet») has three origins: its own memory (managed here), inherited memory (from the parent room, read-only), and subscribed expertise (from other teams, with the authority Binding or Advisory).

Brain Store: discover expertise

The Brain Store is available as its own primary entry in the left app navigation, next to Library and Memory. It makes Corp and Public expertise discoverable by title, description, publisher, genre and tags. Featured and recently updated Brains form transparent collections instead of an anonymous list.

You can browse without a target room. To subscribe, select a room; its owners or responsible governance roles may add the Brain. Existing and inherited subscriptions are marked accordingly.

The origin label distinguishes Corp Brains from this installation from Public Brains. This release provides the local Corp Store: offers, access requests and subscriptions inside your installation. External acquisition and payment are outside that remit; their operating model is the subject of an open decision (BDR-029).

Manage: everything in one place

«Manage» on a memory opens a dedicated surface with everything that belongs to that memory: authority (for subscribed expertise), visibility, memory owner, the decision maker matrix, and the full dreaming configuration, depending on what applies to that memory. The link only appears when there is actually something to manage there. The memory owner defaults to the room owner. You can delegate it to a member, who then curates this memory’s knowledge (visibility, dreaming, owner handover) without gaining room authority such as the decision maker matrix or subscription authority.

Under Settings → «Memory & dreaming», the same path leads through a room and memory picker. This is useful for platform admins without their own room context.

Cadence applies per memory. The scheduler checks every six hours which memories are due; 6, 12, 24 hours or weekly control the actual execution. “Last run” and “Next run” refer to the full cycle of relations, signals and contradiction checks. Individual manual actions such as “Link now” do not reset this cycle status.

Authority: binding or advisory

Advisory

Hits from this expertise are consulted and cited, without priority over other sources. This is the default when subscribing and the safe choice.

Binding

Hits from this expertise are retrieved with priority and marked to the model as authoritative. There is no technical check of the answer against the expertise.

The authority is chosen before subscribing and can be switched later under «Manage» by the room owner. The basis list of every answer shows the authority per citation.

View content before subscribing

Before subscribing, the detail page shows typed counts and representative titles for decisions, rules, documents and entries. Full texts become available in the room only after subscribing. No black-box subscription.

Room owners switch their room’s own memory between private and public under «Manage». A public memory appears in the Brain Store as expertise, is marked as shared, and other teams can subscribe to it.

Only memories of C1 rooms can be shared. Memories of C2 and C3 rooms stay private; the backend rejects sharing there.

Freshness

Every memory shows its state («As of: …»). If it has not been updated for more than 180 days, it carries a visible warning. Refresh its citations with «Re-sync» in the left surface; the action applies to the memory you currently have open.

Unsubscribe

The room owner can cancel a subscription at any time; afterwards this expertise no longer flows into the room’s answers. Inherited subscriptions are managed at the parent room and cannot be cancelled here.

Hiding inherited memories

Inherited memories come from a parent room and are managed there. You can only cancel them in the origin room (link “Manage in origin room” on the card). As room owner, you can, however, choose “Hide for this room” for an inherited memory: it then no longer feeds content into answers in this room; other rooms are unaffected.

Hidden memories stay visible in the “Hidden (n)” section and can be shown again at any time. Every change is recorded as evidence (brain_visibility_change).

Creating child rooms

Creating a new child room is approval-governed. A platform administrator or the parent-room owner can create it directly; other parent-room members submit a request from the same room control.

When a request is approved, the platform creates the room, its own memory, and the requester as named room owner. The new room receives an explicit snapshot of the parent room’s ADR, BDR, and SDR decision makers. C0, C1, C2, and C3 remain fixed platform data classes. A requested room can carry C0 too; it is decided like any other.

A room’s identifier is set when it is created and remains permanent. If a room was named incorrectly or created in the wrong place, change its display name or archive the room: it then disappears from active selection, the tree and agent routing, while evidence remains findable.

The create dialog also asks for the room’s purpose: a required one-line summary of up to 160 characters, plus an optional longer context. Both are recorded as the room’s first decision and appear under its decisions accordingly: the line becomes the title, the context becomes the body. This line is what agents match against when they look for the right room; before this, a freshly created room had no routing profile at all and agents could not find it.

Requesting a child room carries the same requirement: the purpose travels with the request and becomes a decision on the new room only once it is approved; a rejected request leaves neither a decision nor a room. The purpose is set when the room is created and cannot yet be changed afterwards.

Every room has its own agent door: it decides whether agents can reach the room through MCP, separately from the authorization that applies to people. A closed door does not affect a member’s browser access; only the path agents take is affected. A new room takes over its parent’s door state as a snapshot at creation. A room created under a closed parent therefore starts closed too. The top-level room is reachable, so no room is closed until someone closes one. A later change to the parent does not change existing sub-rooms. After that, the room’s owner changes the door in «Details of this room», and every change leaves evidence naming who made it and from which state to which. The door protects only the access path, not the data: because rooms inherit content downward, a closed room with an open sub-room still passes its knowledge on to agents through that sub-room.

What an arriving agent learns about this room comes from its agent instructions: text the room’s owner writes, handed to every agent that connects. Without instructions of its own, an agent reads only the platform’s own text, nothing that sets this room apart from any other. That only matters while the agent door is open: with the door closed, no agent arrives in the first place, so a missing instruction changes nothing. In “Details of this room”, the “Agent instructions” row shows whether anything is set at all: none, default or client-specific; it links to “Manage instructions”, where the room’s owner writes and changes them.

Both facts also appear in the room header, as two chips beside the room name: whether agents can reach the room and whether it briefs them. This way, you can read both without opening the details. A room counts as briefed only when it has a default instruction that reaches every client. An instruction stored for a single client only does not count, because every other agent would still receive nothing.

Room details flag two states as a warning that names what the state causes, who may act, and the next step. A room without a named owner keeps every power with its role holders, but decisions without a named decision maker fall to them alone and the room’s block cards name no contact; only a platform admin names an owner, who must already be a member of the room. An open agent door without instructions hands arriving agents only the platform text; anyone with owner powers writes the instructions or, as a role holder, closes the door, and everyone else is pointed to the room’s owner.

Room workspace orientation

Library shows one record list in the content area, without a second record tree on the left. New record and Import are in the inventory header. There is still exactly ONE left surface for global navigation; Knowledge sources retains its collections and actions such as Add expertise there in Inventory & upkeep. Library, Knowledge sources and Brain Store are not repeated as a horizontal tab band. Opening a detail shows only the selected object; returning restores the filtered list. Change the active room through the compact room control at the top. The info button opens «Details of this room» with data class, role, responsibilities, governance context and usage.

Inventory & upkeep shows collections and what needs attention. Source connections shows rooms, collections, and expanded contents. Textual type labels distinguish rooms from collections even when names match. Map controls, selection, detail views, and upkeep actions remain available. Set Labels directly on the graph to Selected only (default), All, or None. Selected only names the chosen node or the endpoints of a selected relationship, not every neighbor. None hides even selected labels. Full titles remain available in details and hover tooltips. Find nodes opens a collapsible search with keyboard-accessible results only from the current graph view; active type and focus filters still apply. Selection, filters, zoom, and pan preserve the label mode. Another view or room starts with Selected only. The existing Fit selection, Show whole graph, and Back to last focus commands explicitly control framing.

Each area states its task on the row beneath its header, so you can tell where to read a text, check a decision, or maintain a stock:

Find, read, and edit room records. Include inherited content explicitly.
Inspect relationships between decisions, rules, and documents in the readable room inventory.
Understand owned, shared, and subscribed knowledge collections and their upkeep.
Explore connections from rooms to knowledge collections and their contents.
Brain Store: discover published expertise (browse the offers a room can subscribe to).

Where expertise comes from

Browsing the catalog is free. It needs no room, and browsing on its own authorises nothing.

The moment you add expertise, it lands in the room named in the bar. That is why the button names the same room: a target that was filled in for you should not be something you find out about afterwards.

Adding is a governed act. It leaves a trace, and for some offers the publisher decides rather than you. The offer’s own page tells you which case applies.

The room overview

With no memory selected, the console shows the room’s own overview: how many entries and decisions its memories hold together, and how those memories split into the room’s own, what it inherits, and what it subscribes to. Hidden memories are counted on their own line and left out of every other figure, because a hidden memory contributes nothing to what the room answers with.

Binding and advisory describe subscriptions and nothing else. A memory the room owns or inherits carries no such standing, so a room without subscriptions shows neither figure rather than showing two zeros.

The upkeep section reads this room’s dreaming: how many of its own memories dream, when the most recent cycle ran, when the next one is due, and which memory’s last cycle did not complete. Dreaming is configured per memory under Manage, and the picture across all rooms is under Dreaming.

Needs attention separates what this room can act on from what it merely inherits or subscribes to: its own memories are listed by name, and selecting one opens it. A memory counts as stale when nothing in it has been maintained for a long time, the threshold the overview names on the row itself. The overview names which owner is missing: the row reads no governance owner, the accountable role, which is a different role from whoever is responsible for a source’s content; the memory detail labels the source owner the content owner and the room’s governance owner the governance owner, so the two views name the same dimension. A room with nothing of its own outstanding says so in one line. Stale or ownerless memories this room only inherited or subscribed to are listed separately, for information: they are fixed in their origin room or by the publisher, not here, and for an inherited memory the origin room is a link you can open.

Selecting a memory and opening «Decisions & documents» lists its decisions, documents and knowledge entries. «Read passage» on an entry opens the text itself, together with its source, version, validity and owner, and states whether the entry is advisory knowledge or a binding rule. Where they exist, separate links open the source document and any related decision.

The map: opening a memory

The map shows the room chain and every memory that applies here. Select a memory and choose «Show contents» to open it into the decisions and documents it holds; opening a document shows its entries. Only the memory you opened expands, so the map stays readable in a room with many memories.

What the map draws is decided by the control panel in the rail. Switch the room tiers off and the opened decisions stay; switch the three content tiers off and the room relationships stay. «Reset» closes everything again.