This page is machine-translated from French. Read the French original.

Memoirs

A memory It's a short fact that the assistant remembers about you and reuses in its responses. Without it, every conversation starts from scratch.

Function can be disabled at the instance level. If Réglages → Personnalisation It doesn't offer anything; an administrator has disabled it.


Two litters

Scope Content Limit
Global What applies everywhere — «works in a trust company», «prefers concise answers» 5 memories per user
Workspace This is only valid within a specific context — "case X is processed according to procedure Y"« 20 memories per space

The limits are intentionally low. A memory is injected into each conversation: accumulating hundreds of them would saturate the context and degrade the responses instead of improving them. The constraint forces us to keep only the essentials.

For each message, at most five global memories and five spatial memories The most relevant ones are added to the context.


How a memory is born

You create it

Open the memories panel from the conversation and add a fact, choosing its scope. This is the most direct and controlled method.

The assistant suggests it

If automatic extraction is enabled, background processing periodically examines recent conversations and extracts candidate facts.

The process takes place in two stages:

  1. Observation — Identify, within the exchanges, the factual and lasting elements concerning you. One-off remarks, questions, and the content of documents will not be considered.
  2. Filtering — classify each candidate by scope, eliminate duplicates, merge overlapping information, and retain only what remains useful over time.

What remains is recorded as memory. Each conversation is reviewed only once.


What becomes a memory — and what does not

Selected: role and profession, format and language preferences, tools and working methods, sustainable context of a project.

Not selected: the content of your documents (that's the role of indexing), the questions asked, the information valid on a given day, the assistant's answers.

Examples.

Exchange Potential memory
«"I am a chartered accountant at Martin & Fils."» ✅ Overall: profession and employer
«"Always answer me concisely."» ✅ Overall: Format preference
«"This case is following the simplified audit procedure."» ✅ Space: folder rule
«"What is the balance of account 4000?"» ❌ A question is not a fact
«"The contract expires on March 12th."» ❌ This is in the document, not in your profile

View, edit, delete

The memory panel lists everything that is retained, by staff. You can:

  • to modify the text of a memoir; ;
  • DELETE a memory — the effect is immediate on the following conversation; ;
  • delete all to start from scratch.

Your memories belong to you. They are not visible to other users, nor in their conversations, including within a shared space.


Settings — administrator

Réglages → Personnalisation, two independent switches:

  • Memory activated — without it, no memory is stored or injected, and the panel disappears from the interface.
  • Automatic extraction — only has an effect if the first one is active. Disabling it does not erase anything: existing memories remain, but no new ones are automatically offered.

Recommendation. In a sensitive context, activate memory and Leave automatic extraction disabled. : users retain the benefit of personalization, with complete control over what is retained.


Memories or Second Brain?

Two similar mechanisms, not to be confused.

Memoirs Second Brain
Nature Short facts about you Personal knowledge base
Volume A few units Several hundred notes
Food You, or automatic extraction You, the order /note, the agent
Mobilization Systematic injection Search by relevance
Use «"Remember who I am"» «"Remember what I know"»

The two coexist without hindering each other.