Memory

Memory is what the agent keeps from one conversation to the next: a customer’s preference, a detail already given, knowledge built up over time. Without memory, every conversation starts from scratch.

Where memory is created

You can create memories at three levels, and an agent sees every memory that reaches its level:

Level	Where	Who sees it
Folder	Solution folder → Memory tab	Every solution and agent in the folder
Solution	Settings → Memory	Every agent in that solution
Agent	Agent editor → Memory tab	That agent only

Inheritance flows downwards: a memory created on the folder shows up in its solutions marked as Inherited. That is how several solutions share the same knowledge without duplicating it.

The two types

When you click + New memory, the first step is choosing the type.

Type	What it stores	Use when
Structured memory (key/value)	Data organised into defined fields	Customer name, preferences, flags, registration data
Smart memory (Semantic)	Text and documents, searchable by content	Free-form notes, case history, accumulated knowledge

When in doubt, start with the structured one: it is predictable and easy to audit. The semantic one shines when the content is running text and you do not know in advance which fields will exist.

The fields

Field	What it does
Name	How the memory is identified
Scope	Shared, Individual or Per conversation — see below
Description	Tells the agent what to keep here. The field that most affects quality
Default permission	None, Read only, Write only or Read and write
Inject header into prompt	Puts the memory header into the agent’s context
Inject fields into prompt	Puts the stored values into the context
Allow dynamic fields	Lets the agent create fields you did not foresee
Propagate to all agents automatically	Links the memory to every agent without doing it one by one
Active	Uncheck to disable without deleting

Scope

Scope	Visibility
Shared	Information common to all users, available to every agent in the solution
Individual	Separated per person: each user has their own
Per conversation	Only inside that conversation

Choose Individual for personal data — preferences, address, purchase history. Use Shared for business knowledge, not for people’s data.

Scope and personal data

Shared memory is visible to anyone using the solution. Never keep customers’ personal data there.

Permission

Default permission defines what the agent may do with the memory. Read only is the safe choice when you curate the content and the agent just needs to look it up. Read and write lets the agent learn on its own — and demands a well-written description, or it will store things it should not.

Linking to the agent

If you did not tick Propagate to all agents automatically, open the agent editor, go to the Memory tab and link the memory you want.

The tab shows where each memory comes from — Inherited when it comes from the folder or the solution — along with its type, scope and whether it is active.

When the agent seems to have forgotten

Symptom	Likely cause
The agent remembers nothing	The memory is not linked to the agent, or it is inactive
It reads but never writes	Permission set to Read only
It stores the wrong thing	Vague description; state explicitly what should be kept
One customer sees another’s data	Shared scope where it should be Individual