Account memory
In the web chat, Imaginne can keep short facts about you — preferences, your work context, how you like answers written — and use them in later conversations. That's the account memory: it lives on the server, tied to your user, and applies to all your web chat conversations.
On the local surfaces (Desktop, TUI, VS Code), memory is four Markdown notes per project, on your disk. In the web chat, memory belongs to your account and lives on the server. The two are independent. See Memory.
Open memory
Click Memory in the sidebar. The panel shows, top to bottom:
- Storage used — in bytes, with the number of memories and a token estimate.
- Add memory and the shortcut to memory preferences.
- Suggestions to review (when there are any).
- Saved memories.
Add a memory
- Click Add memory.
- Write what to remember — an objective sentence, such as "I prefer direct, to-the-point answers with code examples."
- Optionally, add a title and pick a category.
- Click Add.
Each saved memory shows its category, its source (you added it, approved suggestion, suggestion, imported), and its size. Buttons on each card let you edit and delete.
Categories
| Category | For what |
|---|---|
| Preference | How you want answers (tone, format, language). |
| Profile | Who you are: name, role, area. |
| Company | Your organization, product, market. |
| Project | Context for recurring work. |
| Routine | Things you always do the same way. |
| Instruction | Standing rules ("always cite the source"). |
| Technical | Stack, conventions, tools. |
| Relationship | People and teams you work with. |
| General | Anything that doesn't fit above. |
Suggestions
When you write something that looks like a lasting fact — "I prefer…", "from now on…", "my name is…", "my company…" — Imaginne may propose keeping it. The proposal appears under Suggestions to review, with two buttons:
- Approve — the suggestion becomes an active memory.
- Dismiss — the suggestion is rejected and won't come back.
Nothing enters memory without your approval, and suggestions nearly identical to existing memories are never created.
Automatic memory
Under Settings → Memory (or the gear icon in the memory panel) you choose the mode:
| Mode | Effect |
|---|---|
| Off | Imaginne doesn't suggest new memories and doesn't use saved memories in answers. |
| Suggest (default) | Detected facts become suggestions for you to approve; active memories are used in answers. |
| Automatic | Aimed at low-risk memories. Today, in practice, they also go through your approval — nothing is stored without your say-so. |
How memory is used in answers
Memory is not dumped wholesale into every message. For each request, Imaginne selects only the memories relevant to that message, within a fixed budget — at most 8 memories and roughly 1,200 tokens. A large memory can't monopolize the context: it enters through the budget, not by merely existing.
Memories that get used are marked as such, which also improves the order in which they're picked next time.
Storage limits
| Limit | Value |
|---|---|
| Total per account | 256 KB |
| Size of one memory | 8 KB |
When storage fills up, you'll see: "Your memory storage is full. Imaginne keeps using existing memories, but won't save new ones until you remove or archive some."
Secrets are never memorized
Before any memory is stored — manual or suggested — the content passes through a deterministic safety filter that rejects credentials and sensitive data: API and cloud keys, tokens (JWT, GitHub, Slack, Imaginne's own), private keys, cookies, assignments such as password: …, and credit card numbers.
That filter runs always, including when you type the memory by hand. When in doubt, it refuses.
Things that change with every task shouldn't become memories — they belong to the conversation. Keep what would still be true a month from now.
See also
Was this page helpful?
Report a problem on this pageDo not send passwords, keys, tokens, or customer data.