Skip to content

Agent Memory

Agent's personal knowledge base for learning and improvement.

What is Agent Memory

Agent Memory is where agents store their own knowledge and experiences. Unlike User Memory (shared across all agents), Agent Memory belongs exclusively to each agent.

Memory Ownership:

Owner Stored As Access Purpose
User owner: "user" All agents Your preferences, requirements, context
Agent owner: "self" Only that agent Agent's learned patterns, experiences

How It Works

Each agent builds its own knowledge base over time. Knowledge stored in one agent's memory stays private to that agent.

Example Flow:

Code Review Agent learns:
"This user prefers type hints"
↓
Stored in Code Review Agent's memory
↓
Only Code Review Agent remembers this pattern
(Other agents don't see it)

How Agents Use Memory

Agents automatically manage their memories during task execution:

Before Tasks: Search memories to recall relevant patterns and experiences

During Execution: Add new learnings and update existing knowledge

After Completion: Store successful approaches and lessons learned

Managing Agent Memory

Via Memory Manager

Open Memory Manager from the workspace sidebar. Select an agent's memory card to view and manage that agent's knowledge.

Available Actions:

Action Description
View See all categories and items for that agent
Add Create new category or add items to agent's memory
Edit Update existing knowledge
Delete Remove items or entire categories
Export Download as JSON, Markdown, TXT, or YAML
Import Upload content with AI parsing

Via Agent Tools

Agents use memory tools during conversations to manage their own knowledge autonomously.

Memory Tools:

Tool What Agents Do
list_memory_categories Explore what categories exist
get_memory Retrieve specific knowledge
add_memory Store new learnings
update_memory Refine existing knowledge
delete_memory Remove outdated information

Search with Context Isolation

The search_memory_with_agent tool runs a separate task to find relevant memories without consuming main conversation tokens. The agent explores categories, retrieves content, and returns a focused summary.

Organization

Agent memories are organized in categories. Category names use lowercase letters, numbers, underscores, and hyphens (2-50 characters).

Common Categories:

Type Examples What Agents Store
Capabilities my_capabilities, my_tools What the agent can do
Patterns task_patterns, workflow_tips How to handle tasks efficiently
Knowledge best_practices, common_issues Domain expertise
Learning lessons_learned, feedback_received Improvements from experience

Storage Format

Each memory item includes:

Field Description Example
item_id Unique identifier 20241128_a3f2
content Memory text (max 300 words recommended) "User prefers snake_case naming"
updated_at Last modified date 2024-11-28

Persistence: Memories survive across conversations. An agent's knowledge accumulates over time.

Export & Import

Export Formats

Download agent memories in four formats:

Format Use Case
JSON Structured data for backup
YAML Human-readable structured format
Markdown Documentation and notes
TXT Plain text for simple viewing

Import with AI

Upload text content and select a model. The AI parses the text, extracts information, and organizes it into categories for that specific agent.

Import Flow:

Upload text → AI analyzes → Extracts information → Creates categories → Stores in agent's memory

Privacy & Access

  • Each agent's memory is private to that agent
  • Other agents cannot access another agent's memories
  • You can view and manage any agent's memory via Memory Manager
  • Agent memories persist until deleted