AI memory vs training
AI memory stores notes and brings them back when they are needed. Training changes the model itself. Memory gives an AI facts about you. Training changes how it works.
What does memory do?
Memory saves short notes, such as a rule that you wrote. The assistant adds these notes to your request. The model stays the same, and it reads the notes each time.
What do connectors do?
A connector lets an agent read your email, calendar, or files. The agent finds text and reads it. It can tell you what an email says. It can miss how you would answer, because that is often not written down.
What does training do?
Training uses examples of your work to change the model. The model learns patterns that no document holds, such as when you break your own rule. These patterns are in the model, so there is nothing to search.
Memory keeps the rule you wrote. Training learns when you break it.
Connectors read what the email says. Training learns what matters to you.
An example
A stranger wants a call. They read your report and have a hard question. Your note says no calls with strangers.
An agent with memory finds the note and asks if it should decline. A model trained on your past decisions saw that you take the call when a person did the reading. So it says yes, and it gives that reason.
We wrote these answers. Your own model will answer you.
When should you use each one?
- Use a connector for facts that change, such as today's calendar.
- Use memory for a rule that you can write down.
- Use training for a habit that you cannot write down.
Can you use them together?
Yes. Training sets how the model works, and a connector brings the facts of today. Read how to train AI on your own work.
We will tell you when our models are ready.