Pinning & Promotion
A pin is a memory entry you have curated by hand: a standing rule ("Always add the id field to every model") or a data mapping ("user.uid → auth.users.id"). Pins are the durable, human-reviewed slice of memory — the part that gets exported into the project and recalled by every AI provider.
Pinning
From a running terminal's right-click menu — or its contextual settings panel — choose Pin as Project Rule… or Pin as Data Mapping…. A dialog pre-fills from any text you had selected; Pin saves it into the project resolved from that session's working directory.
Two guards apply at the moment of pinning:
- The secret guard. A line matching a recognizable credential pattern raises a "Possible secret detected" confirmation naming what matched. Cancel is the default; Pin Anyway is the explicit opt-in for a false positive.
- The commit warning. The dialog says where the line goes: pinned project memory is exported to files meant to be committed and shared. Do not paste secrets.
The Pinned Memory panel
Manage Pinned Memory… — from the same right-click menu, or Settings → Pinned Memory — opens the review surface pinning never used to have: a visible, editable list instead of a one-way append.
The panel lists every rule and data mapping you can see from here — the resolved project's, this workspace's, and the global ones — in two sections:
| Section | Contains | Actions |
|---|---|---|
| Pinned | Confirmed pins — entries you pinned or promoted yourself. | Edit, Unpin, delete. |
| Proposed by agents | Pins your agents have proposed on their own. Shown only when proposals exist. | Promote or Discard. |
Every action applies immediately — there is no Save/Cancel footer on this panel.
Promotion
Nothing an agent proposes is pinned silently. A proposal sits in Proposed by agents until you Promote it (it becomes a confirmed pin, exactly as if you had pinned it yourself) or Discard it. If the entry's content changed between the moment the row rendered and your click, the panel surfaces a notice instead of silently pinning content you never reviewed.
Agents cannot cross this boundary from their side either: a memory_update or memory_delete against a pinned entry is refused with a message telling the agent to ask the human to change it in Settings → Pinned Memory.
Scope filter chips
Three chips filter the lists by scope — Project, Workspace, Global — each with a color dot matching the corresponding Neural Map cluster (teal, yellow, green). The chips are a view filter, not a preference: they reset with the panel.
The panel distinguishes two different kinds of empty honestly: "nothing is pinned yet" (the concept explainer) appears only when no pins exist at all, and when pins exist but your chips hide every one, the panel says so and offers show all scopes instead.
What a confirmed pin feeds
A confirmed project-scope pin is exported into the project's committed memory files — .chainabit/memory/rules.md and .chainabit/memory/data-map.md — and folded into CLAUDE.md and AGENTS.md, so every AI provider working in that project recalls it. Proposals are never exported; only what you have confirmed leaves the store.
Where to go next
- Pinned Memory — the step-by-step getting-started walk.
- Scopes — why the chips offer exactly Project, Workspace, and Global.
- Project Model — the project a pin resolves into.