Memory System

KIRA automatically remembers and utilizes important information.

🧠 Memory Structure

Local File System

All memory is stored as Markdown files on your computer.

~/Documents/KIRA/memories/
├── channels/          # Per-channel conversation history
├── projects/          # Project information
├── users/            # Per-user information
├── decisions/        # Important decisions
└── index.md          # Auto-generated index

👤 Per-User Memory

Personal Preferences

KIRA recognizes each user individually and remembers preferences.

Per-User Work History

Manages each user's work history separately.

Team Collaboration Scenario

Everyone shares conversation context from channel discussions.

📚 Auto Save

Auto Save During Conversation

Important information is automatically saved to memory during conversations.

Auto-saved information:

Project-related discussions
Important decisions
User preferences
Task assignments and schedules

🔍 Intelligent Search

Auto-Reference During Task Execution

KIRA automatically finds and uses relevant memory when performing tasks.

User: Send status update email to project lead

KIRA: [Auto-search memory]
      - Project: Project Alpha
      - Lead: Bob Lee (bob@company.com)
      - Current status: Development phase

      Drafting status update email to Bob...

🔐 Data Security

Local Storage

All memory stored only on your computer
Not transmitted to external servers
Stored as files for easy backup and transfer

🚀 Memory Initialization Guide

Building memory systematically when first using KIRA makes it much more effective.

Step 1: Save Team Member Info

Tag team members in a channel and KIRA auto-collects info:

User: @KIRA Remember our team members
      @John Doe @Jane Smith @Bob Lee

KIRA: [Collect info via Slack MCP]
      - Name, email, Slack ID
      - Profile photo, title (if available)

      Remembered 3 team members!

Later usage:

User: Send email to John

KIRA: [Auto-recognize from memory]
      I'll send email to john@company.com.
      What's the content?

Step 2: Teach Organization Structure

Explain your organization structure and roles:

User: KIRA, remember our org structure
      - CEO: John Doe (john@company.com)
      - CTO: Jane Smith (jane@company.com)
      - Dev Lead: Bob Lee (bob@company.com)
      - I'm on the dev team

KIRA: Got it! I've recorded the org structure.
      [Saved to memory]

Tell KIRA about ongoing projects:

User: KIRA, we're working on "Project Alpha"
      Team: Me and Bob
      Period: 2025-01-01 ~ 2025-03-31
      Goal: New feature development

KIRA: Recorded Project Alpha info!

Step 4: Remember Confluence Docs

Share links and ask to remember important documents:

User: KIRA, read and remember this Confluence doc
      https://confluence.company.com/display/TEAM/API-Guidelines

KIRA: [Read doc via Confluence MCP]
      Read and saved API guidelines doc to memory.

      Key points:
      - REST API principles
      - Naming conventions
      - Error handling methods
      ...

Benefits of Building Memory

Building memory early enables:

Recognize people without tagging
"Send email to John" → auto-recognize email address
Auto-reference project context
Use info instantly without searching docs

🛠️ Memory Management

Edit Memory

Fix incorrect or changed information:

User: KIRA, update John's email
      Old: john@company.com
      New: john.doe@newcompany.com

KIRA: Updated John Doe's email address.
      john@company.com → john.doe@newcompany.com

Delete Memory

Remove unnecessary information:

User: KIRA, Bob left the company. Remove from memory

KIRA: Removed Bob Lee's info from memory.

Query Memory

Check stored information:

User: KIRA, tell me about Project Alpha

KIRA: [Memory search]
      Project Alpha:
      - Team: You, Bob
      - Period: 2025-01-01 ~ 2025-03-31
      - Goal: New feature development
      - Status: In progress

💬 Memory Usage Examples

Once memory is built up, KIRA automatically understands context and works smarter.

Using Team Member Info

User: Ask John if he can attend today's meeting

KIRA: [Auto-search John Doe from memory]
      I'll send a Slack DM to John Doe.

      "Hi John, are you available for today's meeting?"

      Should I send it?

Using Project Context

User: What was decided in last week's meeting?

KIRA: [Search Project Alpha related memory]
      Last week's meeting (Dec 10) decisions:
      1. API design confirmed as REST
      2. 1st milestone extended to Dec 31
      3. Bob reassigned to frontend

Using Organization Structure

User: Who do I need approval from for this?

KIRA: [Reference org structure memory]
      The approver for Project Alpha is
      CTO Jane Smith.

      Should I send an approval request email to Jane?

Using Document Info

User: What was the API error code convention again?

KIRA: [Reference saved API guidelines]
      API error code conventions:
      - 4xx: Client errors
      - 5xx: Server errors
      - Custom codes start from 1000

      (Source: API-Guidelines doc)

🧹 Memory Cleanup

Over time, duplicate or outdated information can accumulate. Use the memory cleanup feature to keep things organized.

Request Memory Cleanup

User: KIRA, clean up memory

KIRA: I'll analyze and organize your memory.

      Cleanup results:
      - Merged 3 duplicate files
      - Moved 2 misclassified files (users/ → tasks/)
      - Deleted 5 outdated temp files

      Total: 97 → 87 files

Cleanup Specific Folder

User: Clean up just the users folder

KIRA: Analyzing users/ folder.

      Issues found:
      - 3 files for John Doe (1 profile + 2 task records)

      Should I move the 2 task records to tasks/ folder?

Memory Cleanup Principles

users/: Keep only 1 profile file per person
channels/: Keep only 1 file per channel
tasks/: Store task records separately
Confirm before deleting important info

💡 Tips for Building Effective Memory

1. Use Slack Tags to Remember Team Members

❌ Less effective:

User: Remember John

✅ More effective:

User: Remember @John Doe. He's a backend developer on the dev team.

KIRA: [Auto-collect info from Slack]
      - Slack ID: U01234567
      - Email: john@company.com
      - Name: John Doe

      Saved John Doe's info!

Benefits of Slack Tags

Tagging auto-collects Slack ID and email, so later "Send email to John" automatically finds the address.

2. Include Relationships and Context

User: KIRA, here's Project Beta info
      - Team: Me, John, Jane
      - John is PM, Jane is designer
      - Our Slack channel is #project-beta
      - Related docs: [Confluence link]

3. Be Clear When Updating

User: John moved from dev team to planning team.
      Update the memory.

4. Periodically Check Memory

User: KIRA, show me the team member info I told you

Memory Quality = Response Quality

KIRA's response quality is proportional to stored memory quality. The more accurate and specific information you provide, the smarter it works.

❓ Next Steps

Chat - Effective communication
Task Execution - Tasks using memory
Proactive Suggestions - Memory-based auto suggestions

Memory System ​

🧠 Memory Structure ​

Local File System ​

👤 Per-User Memory ​

Personal Preferences ​

Per-User Work History ​

🤝 Context Sharing ​

Team Collaboration Scenario ​

📚 Auto Save ​

Auto Save During Conversation ​

🔍 Intelligent Search ​

Auto-Reference During Task Execution ​

🔐 Data Security ​

Local Storage ​

🚀 Memory Initialization Guide ​

Step 1: Save Team Member Info ​

Step 2: Teach Organization Structure ​

Step 3: Share Project Info ​

Step 4: Remember Confluence Docs ​

🛠️ Memory Management ​

Edit Memory ​

Delete Memory ​

Query Memory ​

💬 Memory Usage Examples ​

Using Team Member Info ​

Using Project Context ​

Using Organization Structure ​

Using Document Info ​

🧹 Memory Cleanup ​

Request Memory Cleanup ​

Cleanup Specific Folder ​

💡 Tips for Building Effective Memory ​

1. Use Slack Tags to Remember Team Members ​

2. Include Relationships and Context ​

3. Be Clear When Updating ​

4. Periodically Check Memory ​

❓ Next Steps ​