Platform Overview
cogmem-kos is designed to run in two modes, giving you flexibility between simplicity and scale.Deployment Modes
Enterprise Mode
Best-of-breed databases for production workloads
- Postgres for admin and object storage
- OpenSearch for text search with highlights
- Neo4j for entity graph
- Qdrant for vector embeddings
Solo Mode
Single database for development and small deployments
- SurrealDB provides all capabilities
- Simpler setup and operations
- Perfect for prototyping
Architecture Layers
Key Components
Core Models
Contracts
Abstract interfaces that providers implement:- ObjectStore - CRUD for domain objects
- OutboxStore - Event queue for agents
- TextSearchProvider - Full-text search with highlights
- VectorSearchProvider - Embedding similarity search
- GraphSearchProvider - Entity graph traversal
Agents
Event-driven workers that process data:- ChunkAgent - Splits items into passages
- EmbedAgent - Generates embeddings
- IndexTextAgent - Indexes in OpenSearch
- EntityExtractAgent - Extracts named entities
- WikipediaPageAgent - Builds entity pages
Multi-Tenancy
cogmem-kos is multi-tenant by design:- Every object has a
tenant_id - Queries are scoped to tenant
- Optional
user_idfor user-level isolation
Next Steps
Installation
Install cogmem-kos
Configuration
Configure your deployment