MySpend.info · Help & Guides

AI Bookkeeper Memory

Scope: How the AI Bookkeeper's persistent memory works and how to manage it. Memory lets the AI remember standing instructions across all your sessions — so you only have to say something once.

Quick-start summary

3-minute read

Topic	What you do	Where
Add a memory	Say "remember that…", "always…", or "from now on…" to the AI	AI Bookkeeper chat
View memories	Open AI Settings → Memory tab	Gear icon in bookkeeper
Pin a fact	Click the 📌 icon on a fact row	Memory tab
Disable temporarily	Uncheck the checkbox next to a fact	Memory tab
Delete a fact	Click the trash icon → confirm	Memory tab
Compact	Click Compact when 5+ active facts exist	Memory tab
Restore archived	Expand "Archived (N)" → click Restore	Memory tab

What is memory?

Memory is a set of standing instructions that the AI Bookkeeper carries into every conversation. When you tell the AI something once with language like "always", "from now on", or "remember that", it stores the instruction and applies it automatically — without you having to repeat it each time.

Examples of good memory facts

"Always classify AWS charges as a business technology expense"
"My home office is 25% of the apartment — apply that to all utilities"
"I am a sole proprietor using cash-basis accounting"
"Never classify food purchases as personal — I entertain clients regularly"

Memory stores instructions, not data. If you want to ask about your expenses, balances, or spending trends, just ask the AI — it retrieves that from your expense reports and transaction files directly. You don't need to tell the AI your totals or account names.

Adding a memory

You don't need to open any settings. Just tell the AI in conversation:

"Remember that I always pay January rent in December for tax purposes."
"From now on, classify Uber rides as a business expense unless I say otherwise."
"Never suggest capitalising equipment under $500 — I expense everything."

The AI will call its memory tool automatically and confirm: "I've noted that…" The fact appears in your Memory tab immediately.

What triggers a save: Phrases like "always", "never", "from now on", "remember that", "going forward", and similar standing-instruction language. One-off context ("just this time", "for this question") is not saved.

Viewing and managing memories

Open the AI Bookkeeper, then tap the gear icon (AI Settings) → Memory tab.

At the top you'll see a summary line showing your fact counts:

12 active · 3 dimmed · 2 archived

Controls on each fact row

Control	What it does
📌 Pin icon	Marks the fact as permanent — it never ages out
Checkbox	Enables or disables the fact without deleting it
Trash icon	Deletes the fact (requires inline confirmation)
`dimmed` chip	Indicates the fact is in the dimmed state
`pinned` chip	Indicates the fact is pinned (amber highlight)
`from [Name]` chip	Indicates the fact was suggested by an advisor

Memory lifecycle: how facts age

Facts automatically move through three states based on how recently they were relevant to your conversations.

Active

The default state for all new facts. Active facts are injected into every bookkeeper conversation, regardless of what you're asking about.

→ After 60 days without being referenced, an active fact becomes Dimmed.

Dimmed

Dimmed facts are shown at reduced opacity in your Memory tab with a (dimmed) label. They are only included in a conversation when they are relevant to your question — determined by a semantic similarity comparison between the fact and your message. This keeps the AI focused: a standing instruction about rent won't be injected into a conversation about travel expenses.

→ Accessing a dimmed fact (by asking something relevant) returns it to Active. → 90 days after the fact was originally created, a dimmed fact becomes Archived.

Archived

Archived facts are never injected into conversations. They appear in a collapsible "Archived (N)" section at the bottom of the Memory tab. Click Restore on any archived fact to return it to active status. Archived facts are kept indefinitely — they are not deleted automatically.

Pinning: permanent memories

Click the 📌 icon on any fact to pin it. Pinned facts:

Are always injected into every conversation (immune to aging)
Never demote to dimmed or archived, no matter how long since they were last relevant
Show a filled amber pin and a pinned chip

Use pinning for critical standing rules that should never fade:

"I am a freelancer — all income must be reported on Schedule C."
"Home office deduction applies: 20% of rent and utilities are business expenses."

To unpin, click the pin icon again. The fact returns to normal aging.

Pinned but disabled: Unchecking a pinned fact (via the checkbox) still prevents it from being injected — the enabled toggle takes precedence over pinning. This lets you temporarily mute a pinned fact without removing the pin.

Enabling and disabling facts

Each fact has a checkbox. Unchecking a fact silences it — it won't be injected into conversations — but it is not deleted. This is useful for seasonal rules:

Example: "Treat all travel as personal in Q4" — uncheck this fact in Q1–Q3 and re-check it in October.

Re-checking re-enables the fact immediately.

Deleting a fact

Click the trash icon on a fact row. A confirmation prompt appears inline — click Delete to confirm or Cancel to dismiss.

You can also ask the AI: "Forget the fact about home office" and it will identify and remove it.

Deletion is permanent. Deleted facts cannot be restored. If you want to silence a fact temporarily without losing it, use the disable checkbox instead.

Compaction: merging redundant facts

Over time you may accumulate multiple facts that say similar things:

"Always classify Amazon as business"
"Amazon purchases are business expenses"
"Mark Amazon as a business expense going forward"

The Compact button appears in the Memory tab when you have 5 or more active facts. Clicking it asks the AI to merge redundant and overlapping facts into concise replacements. After compaction, a message shows the result:

Compacted: 8 → 5 facts

What gets compacted

Only your own facts (those you added by talking to the AI). Facts suggested by advisors are never compacted — to preserve their source attribution and allow advisors to retract them.

Auto-compaction

If you accumulate 35 active user-authored facts, the next time you ask the AI to remember something, compaction runs automatically before the new fact is saved. You don't need to do anything — the new fact still gets added.

Suggestions from advisors

If you have an advisor (an accountant, bookkeeper, or colleague), they can send you suggested memory facts. When a suggestion arrives, an amber badge appears on the AI Settings gear icon.

Open the Memory tab to see a Pending from advisors section at the top. For each suggestion:

Approve — adds it to your active memory. It appears in the fact list with a from [Name] chip and is used in every conversation going forward.
Reject — dismisses it without saving. The advisor can see that it was rejected.

Advisor-sourced facts behave like any other active fact — they age normally, can be disabled or deleted, and are never auto-compacted.

For more on setting up advisor relationships and sending suggestions, see the Advisors and Clients guide.

What this guide does NOT cover

Out of scope

Advisors and the Clients tab — how advisors send suggestions, manage a suggestion library, and retract suggestions. See the Advisors and Clients guide.
Asking the AI about your data — memory stores instructions, not data. Use the bookkeeper chat to query your expenses and transactions directly.
Export or backup of memory — memory facts are stored in your account and are not currently exportable as a standalone file.
Sharing memory across multiple accounts — each account has its own memory. Advisors can push suggestions, but there is no automatic sync between accounts.