Nia documentation · Latest
Memory
Let Nia remember personal project, list, section and store choices across Tasks, Shopping, BrainDump and recipe transfers.Memory is the common destination-learning system in the Nia V4. It works without an LLM and does not require BrainDump access.
It remembers where you deliberately put familiar tasks and shopping items, then fills reliable destinations automatically the next time. It does not generate the item or decide whether you should buy or do it.
Where it is used
| Workflow | Destinations Memory can help fill |
|---|---|
| Tasks entry and editing | Project and section in the relevant personal workspace |
| Shopping entry and editing | Shopping list, section and stores |
| BrainDump review | Tasks or Shopping destinations, including corrected candidate module choices |
| Recipe ingredient transfer | Shopping list, section and stores for the selected ingredients |
These use one personal system. A deliberate destination correction in a recipe transfer can help the next ordinary shopping entry, and a Shopping choice can help a later BrainDump candidate.
What happens when a match is found
Suppose you have deliberately placed “Butter” in the Chilled section at Green Market several times. When you enter it again and the evidence is reliable:
- Nia fills the remembered destination fields automatically.
- The existing Quick Add-style badges show the applied values.
- A sparkle icon, also used by BrainDump and Memory settings, marks the remembered value, with From memory as its tooltip and accessible name.
- You can keep the values or change them before saving.
There is no extra “accept suggestion” action. A field you explicitly selected, or set through recognized Quick Add syntax, takes priority over an inferred value.
The result can be partial. Nia may know the section but not a store, or a project but not a task section. A weak field is left undecided rather than filled merely because another field was reliable.
A remembered destination must still be accessible. Deleted sections, inactive stores and containers you can no longer edit are not valid destinations.
What teaches Memory
Deliberate choices teach the corresponding fields when the content operation is saved or synchronized:
- explicitly selecting a project or shopping list;
- choosing or correcting a section or store;
- moving an existing task or item to another destination;
- correcting destinations or module choices in BrainDump review;
- choosing destinations when transferring recipe ingredients.
Each field is independent. Moving an item to Chilled does not silently clear its remembered store. Explicitly choosing no section or no store is also a real choice.
A value that Nia filled from memory does not teach itself again merely because it was shown or left unchanged. Changing quantity, notes, sort order or completion does not count as a destination correction.
If you deliberately send the same product to conflicting destinations in a single recipe transfer, the conflicting fields are not treated as a clear new preference.
Correct a remembered destination
- Enter the task or product in the normal workflow.
- Inspect any fields marked From memory.
- Change the field that is wrong, leaving other correct fields as they are.
- Save the task, item, reviewed candidate or ingredient transfer.
The correction becomes part of your personal evidence. Conflicting choices can make a destination uncertain until there is a clear preference; an old destination is not guaranteed to disappear after every single observation.
There is no individual remembered-entry editor in the current settings screen. Use explicit choices in the content workflows to teach corrections, or clear all learned assignments.
Different wording and typos
Matching uses the words and destinations in your own history, rather than a preinstalled product, adjective or synonym dictionary. It considers exact names, complete learned phrases and consistent word evidence across your saved choices.
A longer name can reuse a known phrase when the personal evidence is clear. However, similar wording is not a guarantee of the same meaning. Conflicting or insufficient evidence leaves fields unfilled. More specific explicit history takes precedence over a broader guess.
A conservative fallback can recognize a small typo in a sufficiently long learned word when there is a unique, reliable match. It deliberately avoids short or ambiguous words. This helps route the entry; it does not rename your typed text or create a corrected alias.
Memory does not promise arbitrary synonym understanding, translation or knowledge about food and chores. “Wash laundry” and “do laundry” are not understood as equivalent merely because a human would connect them. Your history must supply enough discriminating evidence.
Product-name matching is separate from unit normalization. A misspelled unit is not corrected by Memory, and route inference does not merge shopping products with different names.
Settings and forgetting
Open your account settings and find Memory.
- Remember preferred destinations enables or disables the shared system.
- The count shows saved assignments for your account.
- Delete remembered assignments clears the common Tasks and Shopping memory.
Disabling Memory also deletes its saved destinations. Reenabling starts learning again; it does not restore the deleted history. Clearing memory does not delete tasks, shopping items or recipes.
Older queued feedback cannot silently recreate the pre-reset history after a reset or disable/reenable cycle. New deliberate choices can teach new assignments while learning is enabled.
Privacy and upgrades
Memory belongs to your account, not to a shared list, cookbook, project or household. Other participants do not receive your personal learning history. Its cache is scoped to the signed-in account and server.
Learning stores task or product naming evidence and destinations, not audio recordings or transcripts. Optional BrainDump requests have their own provider privacy considerations; see BrainDump.
During the V4 upgrade, existing BrainDump destination learning is migrated into the common system. Historical word-level evidence remains historical word-level evidence; Nia does not invent original full task titles that were never stored. Later corrections can update the migrated evidence used by a match.
Previously synchronized personal memory can help supported offline entry. A fresh choice is confirmed with its saved content when synchronization succeeds. Resetting or changing Memory settings requires reaching your server.