Agentic Memory
Agentic Memory provides persistent semantic memory for AI agents.
It is designed to let agents store meaningful information and recall it later based on relevance and intent, instead of relying only on short-lived context windows.
Overview
Many AI agents treat each interaction as an isolated event. Agentic Memory introduces persistent semantic memory so agents can retain useful context across sessions and recall it later based on relevance.
By externalizing memory into a managed long-term store, Agentic Memory helps agentic systems accumulate understanding over time and reconnect past knowledge with new situations.
Core Concepts
A Memory Space is the top-level container for an application, product, or use case.
Key characteristics:
- Groups related Memory Stores under a single context.
- Is scoped to a tenant and Business Group.
- Manages API access at the Memory Space level.
- Is created and deleted through Cloud Portal.
- Deleting a Memory Space removes all associated Memory Stores and memories.
A Memory Store is a collection of memories within a Memory Space. It is typically scoped to an entity such as a user, agent, session, or workflow.
Key characteristics:
- Contains stored memories.
- Supports adding, retrieving, listing, searching, and deleting memories.
- Can be created and deleted by authorized clients using Memory Space credentials.
A Memory is a single stored text entry.
Key Capabilities
Agentic Memory supports:
- Creating and deleting Memory Stores within a Memory Space.
- Adding memories as plain text.
- Retrieving individual memories.
- Listing memories.
- Searching memories using semantic similarity against a text query.
- Deleting memories when they are no longer relevant.
Embedding, indexing, and query optimization are handled by the product.
Limits
| Limit | Default |
|---|---|
| Memory Stores per Memory Space | 100 |
| Memories per Memory Store | 1,000 |
Custom limits can be configured per instance for larger-scale or specialized use cases. Contact your designated Service Delivery Manager for more information.
Typical Use Cases
Use a single Memory Space for the application and create a dedicated Memory Store for each user.
This pattern can store:
- User-specific preferences.
- Past interactions.
- Long-term context.
Benefits:
- Isolates memory between users.
- Supports personalized responses based on historical context.
- Simplifies memory lifecycle management when users are created or deleted.
Use Memory Stores for workflows, agents, or processes.
Examples:
- A customer support automation system with one Memory Store per queue or issue type.
- An autonomous agent system with one Memory Store per agent role, such as researcher, planner, or executor.
Memories can include resolved issues, learned patterns, summaries, and decisions.
An AI booking assistant can use one Memory Store per user inside a shared application Memory Space.
Examples of stored memory:
- Preferred airlines, hotels, or seat types.
- Past destinations and recurring travel patterns.
- Loyalty program memberships.
- Budget ranges or preferred travel times.
- Notes from previous conversations.
The assistant can retrieve relevant memories during each interaction, add new or updated memories, and preserve context across sessions, devices, and long periods of inactivity.
Summary
Agentic Memory provides a managed foundation for persistent semantic memory in agentic and AI-driven systems. It abstracts embedding, vector storage, and scaling so teams can focus on building agent behavior.