Memory

Memory lets you save important information that persists across conversations. Unlike conversation history that resets with each session, memories are stored permanently and can be recalled by the Agent when relevant.

How Memory Works

Memories are project-specific—each project has its own isolated memory storage. When you switch projects, the Agent accesses that project's memories.

You save a memory
        ↓
Stored in project's memory file
        ↓
Agent can retrieve with memory tool
        ↓
Memory content available in conversation

Memories are not automatically included in every conversation. The Agent retrieves them when relevant to your request.

Creating Memories

Manual Creation

Open Memory Settings

Navigate to Settings → Memory.

Click Add New Memory

Click the Add New Memory button to open the creation dialog.

Enter Memory Content

Write the information you want the Agent to remember. Be clear and specific.

Add Tags (Optional)

Add comma-separated tags to organize your memories (e.g., api, authentication, backend).

Save

Click Save to store the memory.

Agent-Created Memories

The Agent can save memories during conversation when it learns something important:

User: "Our API uses JWT tokens with a 24-hour expiration.
       Always refresh tokens before making authenticated requests."

Agent: "I'll save this to memory so I remember your API authentication pattern."
       [Saves memory about JWT token handling]

The Agent decides when information is worth remembering based on:

Project-specific conventions
Repeated patterns you mention
Important decisions or requirements
Technical constraints you specify

Memory Content Guidelines

What to Save

✓ "This project uses Tailwind CSS with a custom color palette
   defined in tailwind.config.js"

✓ "API endpoints follow REST conventions. Always use
   /api/v1/ prefix for all routes."

✓ "Tests must pass before merging. Run 'pnpm test'
   to execute the full test suite."

✓ "The User model requires email verification before
   allowing login. Check user.verified field."

✗ "Fixed a bug today" → Too vague, no actionable information

✗ "Remember to use good code" → Not specific enough to be helpful

✗ "The entire API documentation..." → Too long, use @mentions for large content

Memory Limits

Limit	Value
Content length	2,000 characters max
Word count	300 words max
Memory count	Varies by plan

Plan Limits

Plan	Maximum Memories
Free	5 memories
Starter	15 memories
Pro	25 memories

When you reach your plan's memory limit, you'll need to delete existing memories before creating new ones.

Managing Memories

Viewing Memories

Go to Settings → Memory. All project memories are listed with:

Content preview
Creation and update dates
Associated tags

Searching Memories

Use the search bar to filter memories by:

Content — Match text within memory
Tags — Match tag names

Editing Memories

Hover over a memory card
Click the edit icon (pencil)
Modify content or tags
Click Save

Deleting Memories

Single memory:

Hover over the memory card
Click the delete icon (trash)
Confirm deletion

All memories:

Click Clear All Memories
Confirm bulk deletion

Deleted memories cannot be recovered. Consider editing instead of deleting if you want to update information.

How Agent Uses Memory

Retrieving Memories

The Agent can access memories using the memory tool:

User: "What conventions does this project follow?"

Agent: [Retrieves memories]
       "Based on your saved memories, this project:
       - Uses Tailwind CSS with custom colors
       - Follows REST conventions with /api/v1/ prefix
       - Requires tests to pass before merging"

When Agent Checks Memory

The Agent typically retrieves memories when:

You ask about project conventions or setup
Starting work on a new feature
You reference something "we discussed before"
Making decisions that might conflict with past choices

Memory Format in Context

When retrieved, memories appear as:

Found 3 memories:

1. "API uses JWT tokens with 24-hour expiration"
   [tags: api, auth] (saved 2 days ago)

2. "Run 'pnpm test' before committing"
   [tags: testing, workflow] (saved 1 week ago)

3. "Database migrations require approval"
   [tags: database, process] (saved 3 days ago)

Using Tags Effectively

Tags help organize memories and make retrieval more efficient.

Tag Strategies

Category	Example Tags
Technical area	`api`, `database`, `frontend`, `auth`
Workflow	`testing`, `deployment`, `review`
Conventions	`style`, `naming`, `patterns`
Priority	`important`, `critical`, `nice-to-have`

Example Tagged Memories

Memory: "Always use camelCase for JavaScript variables"
Tags: style, javascript, conventions

Memory: "Production deployments require PM approval"
Tags: deployment, process, important

Memory: "User.findById() returns null if not found, not an error"
Tags: api, database, gotcha

Best Practices

Be Specific and Actionable

"API rate limit is 100 requests per minute per user.
Implement exponential backoff starting at 1 second."

Clear limits and specific guidance.

"Be careful with the API rate limits."

Doesn't tell the Agent what the limits are or how to handle them.

Include the "Why"

"Use UUID v4 for all primary keys instead of auto-increment.
This prevents ID enumeration attacks and simplifies
database replication."

The reasoning helps the Agent make consistent decisions in new situations.

Keep Memories Current

Update memories when conventions change
Delete outdated information
Review periodically to remove stale content

Organize with Tags

Use consistent tag naming
Apply multiple relevant tags
Create tags for common categories

Memories are stored on your machine and not synced to the cloud. If you switch computers, memories won't transfer automatically.

On this page