Skip to main content
Agent memory controls what your agent remembers across conversations. Configure these settings in the Memory tab of your agent’s Workbench.
Agent Memory Settings

​
Three Types of Memory

  1. Session Storage - Conversation history within a thread (session_id). Multiple users can share the same thread.
  2. User Memories - Personal facts about each user (user_id) that persist across all conversations.
  3. Agent Memories - Global knowledge that applies to all users.
Recommended setup: Session Storage (3-5 runs) + User Memories with agentic management enabled.

​
Session Storage

Controls how many conversation exchanges the agent sees.
  • 3 runs = Last 3 back-and-forth exchanges (6 messages)
  • 10 runs = Last 10 exchanges (20 messages)
Terminology: “No. of history runs” (UI) = num_history_responses (code) = max_runs (override)

​
User Memories

Stores personal facts about each user across all conversations. Agentic managed:
  • ✅ Enabled - Agent automatically creates/updates memories
  • ❌ Disabled - You manually create memories via API
Privacy: User memories store personal data. Never store passwords, API keys, SSNs, or credit cards. Memories persist until explicitly deleted.

​
Agent Memories

Stores global knowledge that affects all users. Agentic managed:
  • ✅ Enabled - Agent learns patterns automatically
  • ❌ Disabled (Recommended for Production) - Only admins create memories
Risk: One user’s interaction can change behavior for everyone. Not recommended for production, compliance-heavy, or multi-tenant systems without weekly review and rollback strategy.

​
Max Tool Calls from History

Limits tool calls in context. Set to 0 (disabled) for most cases. Enable only when hitting token limits.

​
Session Summaries

Generates metadata for monitoring/observability dashboards. Does NOT affect context window size.

​
Tool Calls Compression

Compresses verbose tool outputs after N calls. Set threshold to 3-5 for tool-heavy agents.

​
Usage

Memory settings from Workbench are applied automatically. 
from xpander_sdk import Backend, Configuration
from agno.agent import Agent

backend = Backend(configuration=Configuration(api_key="<your-key>"))
agent = Agent(**backend.get_args(agent_id="<agent-id>"))

agent.run(
    "Your message",
    user_id="user@example.com",     # User memories
    session_id="thread-123"          # Session history (thread)
)
Multi-user threads: Multiple users can share the same session_id (e.g., support tickets). Each user keeps their own memories, but sees the same thread history.

​
Debug Logs

Enable debug mode to see memory behavior: 
agent.debug_mode = True
You’ll see: 
DEBUG Added tool update_user_memory
DEBUG Getting messages from previous runs: 6
DEBUG Adding 6 messages from history

​
Troubleshooting

IssueFix
Agent doesn’t remember previous messagesPass same session_id for all messages. Increase history runs to 3+.
User memories not workingPass user_id consistently. Enable User Memories toggle.
Hitting token limitsReduce history runs, enable tool compression, or manage memory programmatically.