What to save to memory, and what to leave out

Memory works best when it captures stable context: the details you want Babbily to reuse across many chats, not just the one you’re in. A little judgment about what goes in keeps your answers better and your memory tidy.

Good things to save

These kinds of context tend to improve many replies over time:

• Communication preference. “Remember that I prefer concise answers with clear next steps.”

• Role or team context. “Remember that I manage customer operations for a B2B software company.”

• Recurring project context. “Remember that Project Atlas is our customer onboarding refresh.”

• Writing style. “Remember that customer emails should sound warm, direct, and low-jargon.”

• Units or formatting. “Remember that I prefer dates in Month Day, Year format.”

• Ongoing audience context. “Remember that my weekly updates are written for non-technical executives.”

Keep these in the current chat

Some details are useful once but shouldn’t become long-term memory: a one-time deadline, a temporary draft direction, a single customer’s private details, a file-specific instruction, or a preference that only applies to this phase of a project. Say so in the chat instead.

“For this chat only, write the proposal in a more formal tone.”

Never save sensitive information

Memory is designed for useful context, not secrets. Don’t ask Babbily to remember:

• Passwords or verification codes.

• API keys or private keys.

• Payment card numbers.

• Government identifiers.

• Raw connected-app credentials.

• Confidential details you aren’t allowed to use in Babbily.

In Settings > Memory, Babbily refuses instructions that mention a password, PIN, API key, card number or ID number, even when the secret itself isn’t included. You’ll see “Memory can’t store credentials or identity secrets.”

Update memory when things change

When a preference or project changes, say so directly.

“Update my memory: for customer updates, use a shorter summary first and put details under headings.”

You can also type the same kind of instruction in the Tell Babbily what to remember or forget… box in Settings > Memory. Write it in English, Spanish, French, German, Chinese, Hindi, Arabic, Portuguese, Russian or Japanese, and Babbily saves the fact in the language you used. The same page lets you review your Memory Profile and remove eligible memory citations.

Check what Babbily remembers

Settings > Memory shows your personal Memory Profile, citations and anything marked Needs review. Project memory is reviewed inside its project. You can also just ask:

“What do you remember that might affect how you answer my work questions?”

If something is outdated, ask Babbily to forget it, and name the exact item. “Forget that Project Atlas is my top priority” works, but a vague “forget that” doesn’t tell Babbily which item you mean. In Settings > Memory, each instruction forgets at most one saved item.

Wait for the confirmation that a save or removal worked. Forgetting a memory doesn’t delete the chat where you first shared it.

A quick quality check

Before you save something, ask yourself:

• Will this still be true in future chats?

• Would it improve many replies, not just one?

• Is it safe to reuse?

• Is it phrased clearly?

• Would I be comfortable seeing it in Settings > Memory?

Where to learn more

The Memory page in the docs explains how to review, update and remove memory, including how forgetting an item relates to the chat where you first shared it. Project memory lives with its project, so review it there. The data controls FAQ covers how memory fits alongside the other privacy controls for sharing, plugins and Library items.

Memory is one of several ways to give Babbily context. Instructions and knowledge files in a Project, or a file attached to a single chat, are better homes for details that belong to one piece of work.

Read the full docs ↗