Skip to main content

Additional features: Memory

When to use​

Turn on Memory when users repeat preferences or account details, for example:

  • Preferred language/tone, summary length, output format (bullets, JSON)
  • Department, location, time zone, frequent request types
  • IDs/usernames that help tools run (employee ID, ticket queue, project code)

Avoid using Memory for secrets, OTPs, or rapidly changing data.

Enable & open settings​

  • Open your agent → Edit.
  • In Additional features, click the gear icon next to Memory.
  • Configure the options below and close the drawer.
  • Publish to apply.

Options explained​

Include memory prompt​

Tell the agent what kinds of facts to store. Example: “Store user’s preferred language, summary length, department, and employee ID.”

Exclude memory prompt​

Tell the agent what must never be stored. Example: “Do not store passwords, OTPs, access tokens, payment info, or medical details.”

User memory (toggle)​

Saves the user’s stable preferences (tone, language, recurring details) and surfaces them in future chats with that same user.

Agent memory (toggle)​

Lets the agent remember helpful patterns from prior outcomes (e.g., which tool worked best for this user’s requests) to improve future answers.

Clear all memory​

Erases all stored memories for this agent. Use when policies change or upon user request.

Good starter policies​

Include​

“Remember: preferred language, reply style (concise/bullets), usual approval chain (manager name), department, location/time zone, employee ID, and commonly used project or ticket queues.”

Exclude​

“Never remember: passwords, OTPs, API keys, access tokens, credit card numbers, sensitive personal/health data, or anything marked ‘temporary’ or ‘one-time’ by the user.”

Prompting tips (tell users what to say)​

  • “Remember my employee ID is E-9142 and prefer Hindi replies.”
  • “From now on, summarize in ≤5 bullets and add a TL;DR.”
  • “Forget my old project code; use ACME-42 instead.”

Best practices​

  • Keep memories short and structured (IDs, booleans, short strings).
  • Use Exclude to enforce your privacy policy.
  • Confirm before storing potentially sensitive items (“Should I remember your employee ID?”).
  • Pair with Context window configuration—use Memory for stable facts; keep the window free for active task context.
  • On request, use Clear all memory or add a “forget” command in your agent’s instructions.

Troubleshooting​

  • Agent doesn’t recall a preference: Ensure User memory is ON and the preference was stated clearly (e.g., “Remember …”).
  • Agent stored something it shouldn’t: Tighten the Exclude prompt and clear memory.
  • Behavior seems inconsistent across users: That’s expected—memory is per end user, not global.