Author shared memories
Every team has small, durable facts the agent should already know. "Our fiscal year ends June 30." "Line 3 runs at 80% capacity in winter." "Customer Acme uses a different SKU naming convention." Shared memories put those facts in front of every Project, App, and Data Collection in your workspace.
You are not training the agent. You are encoding what your team has agreed is true.
Audience: builders
Find the Memories tab
Open the Shared Context group in the left nav, then pick Memories. The page header is "Shared Context"; the sub-tabs underneath select scope.
You will see two scope tabs:
- Team. Memories that apply to everyone on your active team.
- Workspace. Memories that apply to every team in the tenant.
A small badge on each tab counts proposals waiting for review.
Read the two states
Every shared memory lives in one of two states.
- Live. Published and injected into agent context when relevant. Renders with a tag list and an "Updated 5d ago" stamp on the row.
- Pending review. A proposal an analyst or the agent surfaced through chat. Renders under a Pending proposals section with the Pending review badge and a "Proposed 2h ago" line.
Live memories carry a Manual provenance chip in the drawer when an admin authored them, or a Learned chip when the agent proposed and an admin approved them. The chip is your at-a-glance answer to "how did this memory get here?".
Approve a candidate
Candidates appear in the Pending proposals list whenever an analyst correction in chat surfaces something worth keeping, or when the agent itself proposes a memory mid-conversation. Approval is the moment a candidate becomes part of every future answer.
Open the proposal
Click the row under Pending proposals. The drawer shows the proposed body, tags, scope badge, and a "Proposed" timestamp.
Read what would change
For an add or update, the drawer renders the proposed markdown. For a delete proposal, the drawer says it plainly: approving will remove the live memory.
Approve, reject, or withdraw
Workspace admins approve workspace candidates. Active-team admins approve team candidates. The person who proposed the memory can withdraw their own proposal. Each action ends with a toast and the proposal leaves the queue.
Who can approve. Workspace scope: workspace owners and admins. Team scope: admins of the active team. Members can view the list and propose, but only admins act.
Author a memory directly
When you already know the fact, write it yourself instead of waiting for a candidate to emerge.
Open the create drawer
Click the Memory button in the section header. The drawer opens with a blank form titled "New team memory" or "New workspace memory" depending on which scope tab you are on.
Fill the form
- Title. Searchable and shown in the row list.
- Body. Plain language. Markdown is supported: headings, bullets, inline code.
- Tags. Topics this memory relates to. Helps search and grouping.
- Advanced > Valid until. Optional auto-expiry date. Leave blank to never expire.
Save
Click Save memory. The memory lands live in the scope you opened the drawer from.
Edit or retire a live memory
Open a live row to see the rendered markdown, tags, and any expiry date. Admins see two actions in the drawer footer:
- Edit. Opens the same form with the existing values. Save commits a new version. The relative timestamp updates.
- Delete. Asks for inline confirmation in the footer, then retires the memory. A confirmation toast names what was deleted.
The Audit ledger captures every approve, edit, and delete. Filter /audit by the memory you care about when you need the receipt.
Project memories versus shared memories
A project memory belongs to one Project and never leaks. A shared memory is available to every Project, App, and Data Collection in scope. Keep something in the Project when it is specific to one workflow; lift it to shared when more than one Project benefits or the fact is about how your company refers to a concept. See Shared versus project scope for the heuristic and Project memories for the per-Project flow.