Skip to main content
Cognitive Memory is Kubiya’s built-in knowledge management system that provides AI agents with persistent, searchable memory. It enables agents to store context, recall relevant information, and share knowledge across your organization—automatically and without configuration. Cognitive Memory Overview

​
What is Cognitive Memory?

Cognitive Memory is a semantic knowledge base that stores and retrieves information using AI-powered embeddings. Unlike traditional databases that rely on exact keyword matches, Cognitive Memory understands the meaning and context of your data, enabling natural language search and intelligent recall. Key Capabilities:
  • Semantic Search - Find information using natural language queries, not just keywords
  • Temporal Memory - Track and recall information across time and sessions
  • Knowledge Extraction - Automatically identify entities and relationships from unstructured data
  • Cognitive Insights - Generate patterns, connections, and insights from stored knowledge

​
Why Cognitive Memory Matters

Traditional AI agents are stateless—they forget everything after each conversation. Cognitive Memory transforms your agents into learning systems that:
  • Remember across sessions - Agents recall past interactions, decisions, and solutions
  • Share team knowledge - What one agent learns, all agents in the same environment can access
  • Make context-aware decisions - Agents use historical context to provide more accurate, relevant responses
  • Reduce redundant work - Agents don’t re-solve problems they’ve already handled
  • Build organizational knowledge - Every agent interaction adds to your team’s collective intelligence

​
How It Works

Cognitive Memory uses a multi-layered architecture to store and retrieve knowledge. The diagram below illustrates the complete workflow from data ingestion through the knowledge graph to AI-powered retrieval: Kubiya Cognitive Memory Architecture - Ingestion, Graph & Retrieval Workflow Architecture Components:
  1. Data Ingestion & Processing: Raw data from files, databases, and APIs is processed through chunking, embedding, and indexing pipelines, then organized into structured datasets.
  2. Cognitive Memory (Knowledge Graph): The active memory system stores:
    • Entities (Services, Resources)
    • Relationships (Dependencies, Flows)
    • Context (Variable Values)
    • Historical Patterns (Incident Fingerprints)
    • Insights (Derived Knowledge)
  3. AI Agent Search & Memory Tools: Agents use built-in search tools for recall and semantic search, along with a store memory function to capture new insights, creating a continuous feedback loop for learning.
Technical Stack:
  1. Storage Layer: PostgreSQL + pgvector for embeddings, Neo4j for knowledge graph
  2. Embedding Engine: LiteLLM converts text into semantic vectors
  3. Knowledge Graph: Background processing engine extracts entities and relationships
  4. Search Layer: Vector similarity search + graph traversal for intelligent recall
The Flow:

​
Built-in Agent Capabilities

Every AI entity (agents, teams) automatically inherits cognitive memory capabilities. No configuration required.

​
Available Operations

All agents can:
  • store_memory(content, metadata) - Store context with custom metadata
  • recall_memory(query, limit) - Semantic search across stored memories
  • list_memories() - List all stored memories for the current context
  • get_dataset_info() - Get information about the current dataset

​
Environment-Based Datasets

By default, agents automatically use a dataset named after their execution environment: 
Environment: "production" → Dataset: "production"
Environment: "staging" → Dataset: "staging"
Environment: "dev" → Dataset: "dev"
Benefits:
  • Automatic isolation - Production agents can’t access staging memories
  • Shared team context - All agents in the same environment share knowledge
  • Zero configuration - Works out of the box, no setup needed
  • Environment-specific learning - Each environment builds its own knowledge base

​
Agent Integration

Agents automatically inherit cognitive memory capabilities, enabling them to store context, recall information, and coordinate with other agents. The memory operations happen at three key stages:
  1. Pre-Execution - Agents recall relevant context before executing tasks
  2. During Execution - Agents coordinate in real-time by checking shared memory
  3. Post-Execution - Agents store learnings for future use
All agents in the same environment automatically share a dataset, enabling seamless knowledge sharing and coordination.
Learn more about how agents use cognitive memory in the Agent Integration guide.

​
Dataset Organization

Cognitive Memory organizes knowledge into datasets—collections of related memories with shared access controls.

​
Dataset Scopes

Datasets support three levels of access control:
ScopeDescriptionUse Case
USERPrivate to individual user/agentPersonal notes, agent-specific learning
ORGShared across entire organizationTeam knowledge, shared procedures, runbooks
ROLEShared with specific rolesRole-specific guidelines, department knowledge
Default for agents: Organization-level scope, enabling team-wide knowledge sharing.
Learn more about dataset organization and management in Datasets.

​
Multi-Tenant Security

Cognitive Memory is built with enterprise-grade multi-tenancy:
  • Organization isolation - Each organization’s data is completely separate
  • User-level filtering - Individual user memories can be kept private
  • RBAC support - Role-based access control for dataset permissions
  • Audit trails - Complete logging of all memory operations with agent attribution
  • Data encryption - At rest and in transit
All operations are automatically scoped to your organization, ensuring data never leaks between tenants.

​
Integration with Context Graph

Cognitive Memory is tightly integrated with Kubiya’s Context Graph:
  • Unified knowledge layer - Memories appear as nodes in the graph
  • Relationship tracking - Connections between memories, entities, and resources
  • Cross-system queries - Query across cognitive memories and integration data
  • Visual exploration - See memory relationships in the Context Graph explorer

​
Getting Started

​
For End Users

Access cognitive memory through the Compose UI:
  1. Search - Use natural language to find information across all memories
  2. Add Knowledge - Store important context, decisions, and procedures
  3. Manage Datasets - Organize knowledge into logical collections
  4. View Audit Logs - Track who stored what and when

​
For Developers

Use the CLI to manage cognitive memory: 
# Store a memory
kubiya cognitive memory store \
  --dataset production \
  --content "SRE runbook: Scale database during peak hours"

# Search memories
kubiya cognitive memory search "database performance issues"

Agent Integration

How agents use cognitive memory to learn and coordinate

Datasets

Learn about dataset organization and scoping

Semantic Search

Understand how semantic search works

CLI Reference

Complete CLI command reference

​
Production-Ready

Cognitive Memory is battle-tested and production-ready:
  • 93% test coverage - Comprehensive automated testing
  • High availability - Distributed architecture with failover
  • Performance at scale - Handles millions of memories with low latency
  • Monitoring & observability - Built-in metrics and health checks
  • Backup & recovery - Automated backups and point-in-time recovery

​
Next Steps

Now that you understand what Cognitive Memory is, explore how to use it:
  1. Agent Integration - How agents use cognitive memory to learn and coordinate
  2. Datasets - Organize and manage your knowledge
  3. Semantic Search - Find information using natural language
  4. CLI Commands - Command-line integration