DEV Community

Cover image for TencentDB Agent Memory Storage: How Short Term and Long Term Memories Are Stored
QAPulse by SK
QAPulse by SK

Posted on Originally published at skakarh.com

TencentDB Agent Memory Storage: How Short Term and Long Term Memories Are Stored

#agentdesignpatterns #agentengineering #agentmemoryarchitect #agentmemorydatabased

TencentDB Agent Memory Storage is the layer that turns an AI agent’s temporary observations into persistent information that can be reused later. If an agent needs to remember a user’s preference, a previous interaction, a project fact, or a useful experience, the system needs more than a prompt. It needs a deliberate way to represent, store, identify, update, and eventually remove that memory.

That distinction matters because storing agent memory is not the same as simply saving chat history.

Consider a QA agent working on a Playwright project. During a debugging session, it may observe hundreds of events:

User reported a checkout failure
Test started
Browser launched
Login succeeded
API request returned 200
Checkout page loaded
Payment button timed out
Authentication state was refreshed
Test passed
Enter fullscreen mode Exit fullscreen mode

Should all of these become permanent memories?

Probably not.

A useful memory system should identify information that can provide future value.

For example:

The checkout tests require refreshed authentication state
before payment validation.
Enter fullscreen mode Exit fullscreen mode

This is potentially reusable knowledge.

The individual browser-launch event usually is not.

That is the fundamental problem TencentDB agent memory storage needs to solve:

How do we convert useful agent experiences into structured, persistent memories without turning the database into an unlimited dump of conversations?

How do we convert useful agent experiences into structured, persistent memories without turning the database into an unlimited dump of conversations?

What Does Agent Memory Storage Actually Mean?

Agent memory storage refers to the process of persisting information that an AI agent may need beyond its immediate context.

A simplified flow looks like this:

AI Agent
   ↓
Observation
   ↓
Memory Candidate
   ↓
Classification
   ↓
Validation
   ↓
Storage
   ↓
Future Retrieval
Enter fullscreen mode Exit fullscreen mode

The important word here is candidate.

Not every observation should become a memory.

For example:

This is why TencentDB agent memory storage should be designed around information value rather than storage capacity.

A database can store millions of records.

That does not mean an AI agent should remember millions of irrelevant events.

What Should an Agent Memory Record Contain?

A useful memory record normally contains more than a piece of text.

At minimum, you may need:

Memory ID
Agent ID
User/Tenant ID
Memory Type
Memory Content
Created Time
Updated Time
Importance
Confidence
Expiration
Metadata
Embedding
Enter fullscreen mode Exit fullscreen mode

A conceptual record could look like this:

{
  "id": "mem_001",
  "agent_id": "qa_agent",
  "tenant_id": "tenant_123",
  "memory_type": "preference",
  "content": "The team prefers API-based test data setup.",
  "importance": 0.91,
  "confidence": 0.96,
  "created_at": "2026-08-14T10:30:00Z",
  "updated_at": "2026-08-14T10:30:00Z",
  "expires_at": null
}
Enter fullscreen mode Exit fullscreen mode

Now the memory is not just text.

It has identity, ownership, meaning, lifecycle, and context.

That additional information becomes extremely important when the agent has thousands or millions of memories.

Why Metadata Matters

Imagine the agent retrieves this memory:

The user prefers concise reports.
Enter fullscreen mode Exit fullscreen mode

That sounds useful.

But which user?

Which project?

When was this preference recorded?

Is it still valid?

Was it explicitly stated by the user or inferred by the model?

Without metadata, the memory system cannot answer those questions reliably.

A stronger representation is:

{
  "content": "The user prefers concise reports.",
  "memory_type": "preference",
  "scope": "user",
  "confidence": 0.93,
  "source": "explicit_user_statement",
  "created_at": "2026-08-14T10:30:00Z"
}
Enter fullscreen mode Exit fullscreen mode

Now retrieval can become much more intelligent.

The agent can distinguish:

User preference
vs
Project preference
vs
Temporary session information
Enter fullscreen mode Exit fullscreen mode

That is one of the most important design principles behind TencentDB agent memory storage.

Relational Storage Gives Memory Structure

One natural foundation for agent memory is relational storage.

TencentDB for PostgreSQL provides PostgreSQL capabilities that can be used to represent structured memory records.

A simplified schema might look like:

👉 Continue reading the full article on skakarh.com →

Originally published at skakarh.com/tencentdb-agent-memory-storage.
Subscribe to QA Pulse by SK —
weekly signal for QA, Test Automation and AI in Software Engineering.

Top comments (0)