Memory and commands¶
What the assistant remembers¶
By default the assistant knows what is in the conversation you are having. It does not carry facts from one chat into the next.
A memory changes that. It is one short, lasting fact that is brought into later conversations for you: your role, the way your team names things, a preference, a rule that always applies.
There are three places to keep a memory:
| Level | Who it applies to | Example |
|---|---|---|
| Chat | this conversation only | "in this chat, assume the 2026 figures" |
| Project | everyone with access to the project | "our client is called Acme internally" |
| Profile | you, in every chat you have | "I am in Luxembourg; answer in metric units" |
Memories never travel between organisations, and a project memory is visible only to the people who have access to that project.
Saving, listing and forgetting¶
Type / at the start of the message box to open the list of commands, then use one of these:
/memory save Our fiscal year starts in April
/memory save I prefer short answers global
/memory list
/memory list project
/memory forget Our fiscal year starts in April
/memory save …stores the fact. Where it lands depends on where you are: in a chat that belongs to a project it becomes a project memory, in an ordinary chat a chat memory. Add the wordglobalat the end to put it on your profile instead, so it applies everywhere./memory listshows what is remembered. Addchat,projectorglobalto see one level./memory forget <text or number>removes one. If your text matches several memories, you are shown the candidates to choose from rather than having the wrong one deleted.
Commands are carried out by zGentic itself, so they are instant and cost nothing — no model is asked to do anything. Anything that is not a recognised command is sent as an ordinary message.
How a memory reaches an answer¶
Before it answers, the assistant is handed the memories that apply to you here: profile first, then project, then chat.
They are kept short on purpose. Memory is for facts that stay true, not for storing documents. When you want the assistant to work from a document, upload the document. When you want it to follow a rule, save a memory.
What makes a good memory¶
Good: "Our support response time is 4 hours." "Call the product X, never Y." "I report to the CFO."
Not worth saving: a whole policy document (upload it instead), a password or any other secret (never store credentials here), or a detail that only matters in the message you are writing now.