KYBERNESIS
DOCUMENTATION
ACCESS_CONSOLE

How Kybernesis Works

Learn how Kybernesis Brain stores, organizes, and retrieves your knowledge.

Table of Contents

Overview

Kybernesis Brain is a cloud-based memory system that helps you store and retrieve information from multiple sources. You can add memories through:

  • Chat messages (Cmd+K)
  • File uploads (PDF, TXT, DOCX, etc.)
  • Connected apps (Google Drive, Notion)
  • API/MCP (programmatic access)

Once stored, your memories are automatically organized, tagged, and connected to make them easy to find later.

Memory Storage

How Memories Are Stored

When you add content to Kybernesis, the system:

  1. Breaks content into chunks - Large documents are split into ~1200 character segments with overlap to preserve context
  2. Creates searchable embeddings - Each chunk is converted into a mathematical representation for similarity search
  3. Extracts metadata - Title, source, file type, and creation date are preserved
  4. Auto-generates tags - AI analyzes content to suggest relevant tags

What's Stored

Each memory contains:

  • Content - The actual text, extracted from documents or chat messages
  • Title - Auto-generated or user-provided
  • Summary - Condensed version for quick preview
  • Tags - Both AI-suggested and user-added
  • Source - Where it came from (chat, upload, Google Drive, Notion)
  • Relationships - Connections to related memories
  • Metadata - Creation date, file type, size, last accessed

Storage Limits

  • File size: Up to 50 MB per upload
  • Supported formats: Text, PDF, DOCX, code files, spreadsheets, presentations
  • Total memories: Depends on your plan (check your account settings)

Search & Retrieval

Hybrid Search

Kybernesis uses hybrid search combining:

  1. Vector similarity - Finds content semantically similar to your query
  2. Metadata filtering - Searches by tags, date, source, tier

This gives you both precise matches and conceptually related results.

Search Options

Via Web UI:

  • Topology graph - Visual search by clicking nodes and following connections
  • Chat interface (Cmd+K) - Ask questions, get relevant memories

Via API:

  • POST /retrieval/hybrid - Programmatic search with filters
  • See API Reference for details

Via MCP:

  • kybernesis_search_memory tool for AI agents
  • See MCP Setup for configuration

Search Results Include

  • Relevance score (0.0-1.0)
  • Content preview or full text
  • Tags for filtering
  • Source and creation date
  • Related memories via graph connections

Automatic Organization

Auto-Tagging

Every 60 minutes, Kybernesis analyzes untagged memories and suggests:

  • Topic tags (e.g., "authentication", "database")
  • Entity tags (e.g., "PostgreSQL", "AWS")
  • Action tags (e.g., "troubleshooting", "configuration")

You can always add your own tags manually via the topology UI.

Relationship Detection

The system automatically finds connections between memories based on:

  • Shared tags
  • Similar content
  • Common entities
  • Same source

These connections appear as links in the topology graph.

Memory Prioritization

Kybernesis tracks:

  • Access frequency - How often you view or search for this memory
  • Recency - When it was created and last accessed
  • Manual pins - Memories you mark as important

This helps surface the most relevant results in searches.

Storage Tiers

Memories automatically move between three tiers based on usage:

Hot Tier

  • Purpose: Frequently accessed, high-priority memories
  • Search priority: Highest
  • Access speed: Fastest

Stays hot if:

  • Accessed within last 3 days
  • High priority score (≥0.65)
  • Manually pinned by you
  • Many connections to other memories

Warm Tier

  • Purpose: Occasionally accessed memories
  • Search priority: Medium
  • Access speed: Normal

Moves to warm if:

  • Accessed within last 21 days
  • Has manual tags
  • Moderate connections

Archive Tier

  • Purpose: Rarely accessed memories
  • Search priority: Lower
  • Access speed: Slower

Moves to archive if:

  • Not accessed for 30+ days
  • Low priority score
  • Few connections
  • No manual tags

Note: Accessing an archived memory automatically promotes it back to warm or hot tier.

Manual Tier Control

You can pin memories to stay in hot tier:

  1. Open memory details in topology UI
  2. Click the pin icon
  3. Memory will remain hot regardless of access patterns

Data Security

Encryption

  • In transit: All API requests use HTTPS (TLS 1.3)
  • At rest: Data encrypted using industry-standard encryption
  • API keys: Stored securely, never logged or exposed

Access Control

  • Organization isolation: Each organization's data is completely separate
  • API authentication: All API requests require valid API key
  • User permissions: Coming soon - role-based access control within organizations

Data Retention

  • Active memories: Stored indefinitely while your account is active
  • Deleted memories: Permanently removed within 24 hours
  • Account deletion: All data deleted within 30 days of account closure

Compliance

  • SOC 2 Type II: Coming 2025
  • GDPR: Data export and deletion available on request
  • Data location: Hosted in US with EU region coming soon

Your Control

You can:

  • Export all data via API or web console
  • Delete specific memories individually or in bulk
  • Disconnect connectors to stop syncing external data
  • Revoke API keys instantly from settings

Next Steps

Support