Using agent memory in conversations

Edit online

Use agent memory to create more efficient and personalized interactions with AI agents by allowing them to remember and use information about you across chat sessions.

When agent memory is enabled, the AI agent automatically learns from your conversations and uses stored information to provide more relevant and contextual responses.

How the agent learns from conversations

The agent extracts relevant information from your conversations and stores it in memory. This process happens automatically as you interact with the agent.

What the agent remembers

During conversations, the agent identifies and stores:

  • Preferences you express about products, services, or activities
  • Personal details you share (such as name, role, or location)
  • Work-related context (such as responsibilities or frequent tasks)
  • Communication preferences (such as preferred tone or level of detail)
  • Plans and intentions you mention (such as upcoming events or goals)

How information is extracted

The agent analyzes your messages to identify facts and preferences worth remembering. For example:

Your message:

I'm based in Bangalore and work as a Customer Success Manager. I frequently book flights to visit customers.

What the agent remembers:

  • Base location: Bangalore
  • Role: Customer Success Manager
  • Frequent activity: Booking flights for customer visits

The agent does not store the entire conversation, only the extracted facts.

Instructing the agent about memory

You can explicitly tell the agent what to remember or forget by using natural language instructions.

Asking the agent to remember information

Use clear instructions to add information to memory:

Remember that I prefer morning flights
Keep in mind that I work in the Pacific time zone

The agent typically responds with a confirmation such as "Got it! I'll keep in mind that you prefer morning flights for any travel‑related suggestions."

Asking the agent to forget information

Use clear instructions to remove information from memory:

Forget my previous role as Customer Success Manager
Remove my flight preference from memory

The agent typically responds with a confirmation such as "Your details have been removed from the record."

Asking the agent to update information

Update existing memory by providing new information:

Update my base location to Kochi
Change my role to Solution Architect

The agent updates the relevant memory items and confirms the change.

Using memory across chat threads

Agent memory persists across all your chat threads and sessions. Information stored in one conversation is available in all future conversations with any agent that has memory enabled.

Example scenario

Thread 1 - Initial conversation:

You: I'm based in Bangalore and prefer direct flights.
Agent: Noted, I'll remember your location and flight preference.

Thread 2 - Later conversation:

You: Book a flight to Mumbai for next week.
Agent: I'll look for direct flights from Bangalore to Mumbai for next week, based on your preferences.

The agent uses information from the first thread without requiring you to repeat it.

Using memory across sessions

Agent memory persists across different login sessions. When you return to watsonx Orchestrate after logging out, the agent still remembers information from previous sessions.

This continuity allows you to:

  • Resume conversations without re-establishing context
  • Maintain consistent preferences across time
  • Build on previous interactions seamlessly

How memory improves interactions

Reduced repetition

Without memory:

You: Book a flight to Delhi.
Agent: What's your departure city?
You: Bangalore.
Agent: Do you have any airline preferences?
You: I prefer direct flights.

With memory:

You: Book a flight to Delhi.
Agent: I'll search for direct flights from Bangalore to Delhi.

Contextual responses

The agent provides responses tailored to your specific context:

You: What's the weather like?
Agent: The current weather in Bangalore (your base location) is 28°C and partly cloudy.

Proactive assistance

The agent anticipates your needs based on stored preferences:

You: I need to travel to Chennai next month.
Agent: Based on your preference for direct flights and morning departures, I found several options from Bangalore to Chennai.

Verifying what the agent remembers

Check what information the agent has stored about you at any time:

What do you remember about me?
What's my base location in your memory?
Show me what you know about my work preferences.

The agent provides a summary of relevant stored information. For a complete view of all stored memory, open the Agent memory pop-up from the left panel.

What to do next