Memory is the organisation model, kept as ordinary docs. Every task reads from it and writes back into it; you can read it, edit it, and reorganise it. This page covers what it holds, how it stays current on its own, how it is organised, and how to correct it when it is wrong.
What memory is
Each directory — one modelled organisation — has its own memory. It holds everything the work has touched:
- The model itself. A doc per stage of the business — market, customers, product, pricing, channels, strategy, team, operations — first written by the company audit from your website, then refined by every task.
- Every dossier. The entry doc and artefacts of every task, filed where they belong.
- Every answer you gave. A data request answered once is written into the work and kept, so no later task asks again.
- The living documents. A summary per entity the work has touched — a customer, a product, a competitor — updated by each run that touches it.
- What you added. Docs you wrote or pasted in, meeting transcripts captured by the Chrome companion, newsletters that arrived in the inbox.
Ask a question in Ask mode and it is answered from memory alone, citing the records it used, with nothing new built.
Memory vs directory settings
Both carry standing context into every task. They are for different things.
If the stage or the site changed, change settings and re-run the company audit. If one doc is wrong, edit it.
How memory keeps itself current
You do not maintain memory by hand, though you can. Four things update it without you:
- The Summary stage of every task rewrites the living document of each entity the run touched, so the customer doc reflects the latest dossier about that customer.
- Answers write through. Answer a data request and the number lands in the artefacts that used the placeholder and in the model.
- The company audit re-reads your website stage by stage whenever you run it, so a repositioning reaches the model.
- Automatic organisation files each finished task's output into the right section. Turn it off in directory settings if you would rather file by hand; Reorganise directory re-files everything in one pass when you turn it back on.
Every change adds a version rather than overwriting. A run that rewrites a doc you had reviewed cannot destroy your copy; the version selector in the doc header opens earlier states. See Docs and versions.
How memory is organised
Memory is organised into sections: the headings you see in the sidebar and under Settings → Memory → Sections. A section is a shelf — Customers, Pricing, Board material — and automatic organisation files new docs onto the right one.
Sections are what you work with. Behind them the system files docs by category and tab as well, which is how the sidebar and the Memory table agree about where a doc lives; you do not need to manage those.
To change the shelves:
- Add, rename or reorder sections under Settings → Memory → Sections.
- Move a doc from its row on the Memory table, or from the doc's own settings.
- Reorganise directory on the Memory table re-files the whole directory in one pass — useful after a long stretch with automatic organisation off.
Reading memory
The Memory page is every doc in the directory in one table, filterable by interface type, origin, section, versions, sidebar placement and dates, with two filters worth knowing: unindexed (docs not reachable from the sidebar) and plain (documents with no interface). See Memory for the table itself.
For "what do we know about X", ask in Ask mode. For "show me everything we have", use the table. For "where did that number come from", open the artefact and read its sources.
Editing memory
Every doc in memory is editable. Open it and change it, the way you would any doc. This is how you correct the model:
A correction in memory holds for every later task. A correction in a request holds for that task alone.
When you correct a doc, say why in the doc. The reviewer stage reads memory too, and a stated reason stops a later run from "fixing" your correction back.
What memory does not do
- It is per directory. Your company, a competitor and a client each get their own; nothing crosses between them unless you copy it.
- It does not copy connected tools. Connectors are read on demand for the task at hand; the conclusions are kept, the raw data is not.
- Files you attach to a task stay with the task. They inform that run and are not written to the model.
Troubleshoot memory
A task ignored a correction I made. Check you edited the doc the work reads — the living document for the entity, not an older dossier about it. The Sources panel on the artefact shows which docs it used.
New docs are landing unfiled. Automatic organisation is off in directory settings, or the section that fits does not exist yet. Turn it on, add the section, then Reorganise directory.
I cannot find a doc. On the Memory table, clear filters and check unindexed — it lists docs not reachable from the sidebar.
Two docs disagree. That is a divergence issue waiting to be raised; run the question on Connected and the reviewer will flag it, or edit the wrong one.