Short-Term Memory (Session Context)¶

Conversation context automatically maintained within each session.

Memory Type Comparison:

Type	Storage Location	Scope	Lifecycle
Short-Term Memory	Conversation Turns	Single Session	Persisted until deleted
Long-Term Memory	Database Categories	Cross-Session	Persisted until deleted

Automatic Management¶

The system automatically manages short-term memory. You don't need to manually operate it.

Each message consumes tokens. Longer conversations use more tokens and may reach the model's context limit.

Token Management:

Aspect	Behavior
Context Window	All turns sent to model as context
Token Limit	Determined by selected model
Long Conversations	May exceed model's context limit
Solution	Use conversation compression

When conversations become long, you can compress them to reduce token usage while preserving key information.

Compression Types:

Type	Method	Use Case
Brute Force Compression	Keep only first and last message of each user-assistant turn	Quick reduction needed
Precise Compression	AI summarizes tool content	Quality preservation important

How to Compress:

Open conversation settings and select compression type. The system compresses tool outputs while keeping user and agent messages intact.

Compression Flow:

Long Tool Output → Compression → Shorter Summary → Reduced Tokens

Conversations persist in the database even after closing them. You can return later to view or continue.

Access past conversations from the conversation sidebar. Select any conversation to view its complete message history.

Available Actions:

Action	Description
View	Read all conversation turns
Continue	Add new messages to conversation
Export	Download as JSON or Markdown
Delete	Remove conversation (soft delete)
Compress	Reduce token usage