Core Concepts

Understanding the Rice architecture.

Rice provides a complete cognitive infrastructure for AI agents. It consists of two main components:

ComponentPurpose
StorageStorage and memory engine
SlateState and runtime context management

Storage

Rice's storage layer is a high-performance database designed for Multi-Agent AI Systems. It handles:

  • Vector Search: Semantic similarity over embeddings
  • Graph Database: Structural relationships between entities
  • Agent Memory: Lightweight scratchpad for multi-agent coordination
  • Access Control: Bitmap-based ACL for zero-latency permission checks
  • Pub/Sub: Real-time notifications with semantic subscriptions

RiceDB uses Hyperdimensional Computing (HDC) for noise-tolerant, high-speed operations on memory traces.

Learn more about Storage →


Slate

Slate is the state and runtime context management layer that sits between your AI agents and storage. It provides a four-component cognitive memory architecture:

Memory TypePurpose
Working MemoryPre-fetches context with decay and attention scoring
Episodic MemoryStores interaction traces for learning from experience
Procedural MemoryExecutes compiled skills server-side
Semantic MemoryStores invariant facts and knowledge

Slate turns stateless LLMs into learning, stateful agents.

Learn more about Slate →


How They Work Together

Your AI Agent
     ↓
   Slate (Memory Management)
     ↓
   Storage (RiceDB)
  1. Your agent calls Slate APIs (focus, drift, commit, reminisce)
  2. Slate manages the cognitive abstractions (attention, decay, traces)
  3. RiceDB handles the low-level storage, indexing, and vector operations

This separation means you can focus on building intelligent agent behavior while Rice handles the memory infrastructure.