About 5 min read

Memory System

Karma One's memory system acts as your Second Brain -- a persistent, private knowledge store that grows smarter the more you use it. Memories carry across every conversation, so the AI always has context about who you are, what you care about, and how you work.

The Second Brain Concept

Traditional AI assistants forget everything the moment a conversation ends. Karma One is different. Its memory system captures and retains the important details from your interactions, building a personalized profile over time.

This means:

  • You do not have to repeat your preferences in every conversation.
  • The AI can reference past projects, decisions, and insights.
  • Your assistant becomes more useful the longer you use it.

Memory Categories

Memories are organized into five categories, each serving a distinct purpose:

| Category | What It Stores | Example | |----------|---------------|---------| | Facts | Objective information about you | "User lives in San Francisco", "User's company is Acme Corp" | | Preferences | How you like things done | "User prefers TypeScript over JavaScript", "User likes concise answers" | | Projects | Context about ongoing work | "User is building a mobile app with React Native", "Project deadline is March 15" | | Insights | Patterns and learnings | "User's API latency issues were caused by N+1 queries", "Team velocity increases with 2-week sprints" | | Processes | Step-by-step workflows | "Deploy process: run tests, build Docker image, push to staging, then production" |

Automatic Memory Extraction

The AI continuously analyzes your conversations and automatically extracts information worth remembering. You do not need to explicitly say "remember this" -- though you can.

What Gets Auto-Extracted

  • Personal and professional facts you share
  • Preferences you express ("I prefer dark mode", "Always use metric units")
  • Project context and decisions
  • Technical insights and solutions to problems
  • Recurring workflows and processes

Examples of Automatic Extraction

You say:

"We decided to go with PostgreSQL for the new project because of its JSON support."

AI remembers: Project uses PostgreSQL; chosen for JSON support.

You say:

"Please always format code examples in Python unless I specify otherwise."

AI remembers: User prefers Python for code examples by default.

Manual Memory Management

You have full control over your memories. You can add, search, edit, and delete them at any time.

Adding a Memory

"Remember that our production server is at 192.168.1.100."
"Save this: the API rate limit is 1000 requests per minute."

Searching Memories

"What do you remember about my project setup?"
"Search your memory for anything related to deployment."

The AI will retrieve and display relevant memories, showing you exactly what it knows.

Deleting a Memory

"Forget my old office address."
"Delete the memory about the Python preference -- I've switched to Rust."

Cross-Session Persistence

Memories persist indefinitely across all your conversations. Whether you start a new chat today or six months from now, the AI will have access to everything it has learned about you.

This is particularly powerful for:

  • Long-running projects where context builds over weeks or months
  • Multiple workstreams where the AI can connect dots between conversations
  • Team onboarding where you can tell the AI about team norms once and have it apply them consistently

How Memory Search Works

When you ask a question, the AI automatically searches its memory for relevant context before responding. This happens transparently -- you do not need to ask it to "check memory."

The search process:

  1. Your message is analyzed for key topics and entities.
  2. Relevant memories are retrieved using semantic search.
  3. Retrieved memories are included as context for the AI's response.
  4. The response is informed by both the current conversation and stored memories.

Tip: You can explicitly ask the AI to use or ignore its memory. Say "Answer this without using your memory" if you want a fresh perspective.

Privacy and Ownership

Your memories are yours alone.

| Privacy Feature | Details | |----------------|---------| | Per-user isolation | Memories are scoped to your account. No other user can access them. | | No training use | Your memories are never used to train AI models. | | Full control | You can view, export, and delete any or all memories at any time. | | Encryption | Memories are stored encrypted at rest. |

Memory in Practice

Scenario: Recurring Weekly Report

Week 1:

"Help me write a weekly status report. Our team uses this format: accomplishments, blockers, next week's plan."

The AI remembers the format.

Week 2:

"Write this week's status report."

The AI already knows the format and applies it automatically.

Scenario: Technical Preferences

"I always use 4 spaces for indentation and prefer functional components in React."

From this point forward, every code example the AI generates will follow these conventions.

Scenario: Cross-Conversation Context

Conversation A: "We're migrating from MySQL to PostgreSQL. The target date is April 1."

Conversation B (weeks later): "How's the database migration going?"

The AI recalls the MySQL-to-PostgreSQL migration and the April 1 deadline without you restating it.

Best Practices

  • Let it learn naturally. You do not need to front-load the AI with information. Share context as it comes up and the memory system will capture what matters.
  • Review periodically. Ask "What do you know about me?" occasionally to audit stored memories and correct anything outdated.
  • Be specific when correcting. Instead of "forget everything about my project," say "forget that the deadline is March 15 -- it has been moved to April 1."
  • Use categories mentally. When sharing information, think about whether it is a fact, preference, project detail, insight, or process. This helps the AI categorize it correctly.