Memory Configuration

4 memory types, namespaces, TTL, and privacy controls

Memory Configuration

The platform provides four memory types that allow agents to retain and recall information across conversations. Memory is scoped by namespace (user, team, agent, global) and subject to privacy controls and TTL policies.

Memory Types

Short-Term Memory (Redis)

Purpose: Context within a single conversation or session.

• Backend: Redis with configurable TTL
• Default TTL: 1-7 days
• Use case: Multi-turn conversations where the agent needs to remember what was said earlier in the session
• Scope: Per-conversation

When to enable: Almost always. Most agents benefit from short-term context.

Long-Term Memory (Qdrant)

Purpose: Facts, preferences, and knowledge that persist across conversations.

• Backend: Qdrant vector database
• Default TTL: 90-365 days
• Use case: Remember user preferences, past interactions, learned facts
• Scope: Per-user or per-agent (configurable via namespace)

When to enable: When agents should "know" returning users and their history.

Graph Memory (Neo4j)

Purpose: Entity relationships and structured knowledge.

• Backend: Neo4j graph database
• Default TTL: 90 days
• Use case: Track entities (people, accounts, projects) and their relationships
• Scope: Per-tenant or per-team

When to enable: When agents need to understand "who relates to what" — customer accounts, org charts, project dependencies.

Procedural Memory

Purpose: Learned task patterns and workflows.

• Backend: PostgreSQL + Vector embeddings
• Default TTL: 60-120 days
• Use case: Agents that improve over time by learning from repeated interactions
• Scope: Per-agent

When to enable: For agents that handle repetitive tasks and should optimize their approach over time.

Namespace Scoping

Memory is organized into four namespaces that control visibility:

Example

User "alice" asks: "Remember that I prefer summaries in bullet points"
→ Stored in namespace: user (only alice sees this)

Team "engineering" discusses: "Our sprint planning is every Tuesday"
→ Stored in namespace: team (all engineers see this)

Agent learns: "Users often ask about vacation policy — search KB first"
→ Stored in namespace: agent (applies to all users of this agent)

Configuring Memory in Agent Studio

Step 1: Enable Memory Types

In Studio > Agents > [Agent] > Memory tab:

Toggle each memory type on/off:

• Short-Term: Enable for conversational context
• Long-Term: Enable for cross-conversation memory
• Graph: Enable for entity tracking
• Procedural: Enable for task learning

Step 2: Configure Extraction

Memory extraction controls what the agent automatically remembers:

Step 3: Configure Consolidation

Memory consolidation merges related entries to prevent bloat:

Step 4: Configure Retrieval

Memory retrieval controls how stored memories are surfaced:

TTL Configuration

Each memory type has a configurable time-to-live per agent:

Navigate to Studio > Memory > TTL Configuration:

Privacy Controls

Memory Consent

Users can control their memory through the Portal > My Data page:

• Opt-out of memory — disable memory collection for their interactions
• View stored memories — see what the agent remembers about them
• Delete memories — remove specific entries or all entries
• Export data — GDPR data export

Memory Purge

Admins can purge memory in Studio > Memory > Purge:

• Filter by agent, memory type, date range
• Requires confirmation before executing
• Creates an audit log entry

Audit Trail

All memory operations are logged:

• Who created/accessed/deleted the memory
• When and from which agent
• The operation type (create, read, update, delete, consolidate, purge)

View the audit log in Studio > Memory > Audit Log.

Monitoring Memory

Dashboard Metrics

In Studio > Memory > Dashboard:

• Total entries by memory type
• Storage usage (MB) per type
• Hit rate — % of retrievals that found relevant memories
• Average TTL — how long entries live before expiring
• Namespace distribution — user vs team vs agent vs global

Graph Visualization

In Studio > Memory > Graph tab:

• Visualize entity relationships stored in graph memory
• Click nodes to see properties and connected edges
• Filter by entity type (person, organization, project, etc.)

Best Practices

1. Start minimal — enable short-term first, add others as needed
2. Set appropriate TTLs — don't keep memories forever
3. Use namespaces wisely — user data stays in user, shared knowledge in team or agent
4. Monitor consolidation — check that merges are accurate
5. Respect privacy — enable consent controls for user-facing agents
6. Test retrieval — verify that stored memories are actually surfaced when relevant
7. Purge regularly — remove stale data to improve retrieval quality

Troubleshooting

Next Steps

• Knowledge Bases Guide — combine memory with knowledge retrieval
• Platform-Managed Agents — configure memory for your agent
• Security & Compliance — memory privacy and GDPR controls