AI memory
The assistant's notebook about your business, and the inbox where it asks before writing in it. A memory is a short rule, preference, decision, fact, open question or pattern ("we always exclude asbestos removal", "Cam prefers totals before detail", "the Smith job's client wants Friday updates"). Anything the assistant works out for itself lands here as a proposal; nothing reaches the assistant until a person keeps it. It is for whoever owns how the business works: an owner, a lead estimator, a PM.
- Admin or team role to open the page. Clients and other roles are sent to the dashboard.
- Permission to change the AI's standing rules (on for staff by default; an admin can switch it off per person under Team modules). Without it you can read but not keep or bin.
- Something to review. The badge on AI Memory in the nav counts proposals waiting.
The four tiles at the top count memories In use (always-on and on-demand), Awaiting you, Never surfaced yet and Retired. Press What the AI is working with right now to expand the always-on list: every memory that goes into every conversation, with its reason and whether it is company-wide or yours. A character meter shows how heavy that list is against the guideline; over it, nothing is dropped, but the panel suggests leaving anything not truly universal for on-demand recall.
- Each card shows the proposed memory, Why the assistant thinks so, its kind, its scope (Company, You or a job's name) and where it came from: Inferred in chat, Noticed across several jobs, Noticed from your edits, Imported from your saved rules, Imported from project notes.
- Press the tick (Keep it — the AI will use this from now on) or the cross (Bin it — the AI will not use this).
Kept memories are live immediately: rules and preferences in every conversation, decisions and facts recalled when relevant. Binned ones are kept as a record so the assistant does not propose the same thing again. There is no confirm step and no edit; if the wording is nearly right, bin it and tell the assistant the right version in chat.
Press All, Rule, Preference, Decision, Fact, Open question or Pattern. Each chip shows its count.
Say it in chat: "remember that we never price demolition, the client's demolisher does it". Something you state yourself is saved as live at once and does not pass through this page. Something the assistant infers from a conversation is saved as a proposal and waits here.
- The assistant proposes a memory when it notices how you work in a conversation. It tells you it has done so and that nothing changes until someone approves it here.
- Imports of your older saved rules and project notes arrive as proposals.
- Prices never get in. Every memory is checked before saving; a dollar figure, a unit rate or a trade's hourly rate is refused and the assistant is told to put it in the rate book instead. The one exception is a threshold that selects a rule ("over $250k contract sum, add the levy"), which names a boundary and prices nothing.
- A memory about a specific job loads only inside that job. Company and personal memories load everywhere.
- The page refreshes itself when the assistant changes memories in chat.
- The tick and cross have no labels. Hover for the tooltip; tick keeps, cross bins.
- Nothing you bin is deleted. It is kept as a record, and the assistant will not propose the same wording again; if it is "genuinely different" it has to say what changed.
- You cannot edit a proposal. Keep or bin, then correct the assistant in chat.
- A rule the assistant learned does not change any price. "Our sparkies are $95/hr and we always run tradesman plus apprentice" becomes two things: a labour rate in the rate book and a rule here.
- Memories are not organised by estimate mode or skill; they are tagged by topic and scope. A pattern can only be company-wide or yours; a decision or open question can only belong to a job or the company.