Core Concepts

Rice is a serverless backend designed around key capabilities that enable stateful, context-aware AI agents.

1. Long-Term Memory (Hyper Vectors & Associations)

To make intelligent decisions, agents need access to vast amounts of persistent knowledge. Rice unifies two powerful paradigms to create a robust Long-Term Memory system.

Hyper Vectors

Enable high-fidelity semantic search. Agents can "recall" information based on meaning rather than just keywords.

• High Dimensionality: Data is projected into thousands of dimensions, capturing subtle nuances in meaning.
• Noise Tolerance: The system is robust to imperfect queries or noisy data.
• Example: Searching for "legal precedent for copyright" matches "IP infringement case law" even without shared keywords.

Neural Associations

Encodes relationships directly into the high-dimensional space. Unlike traditional graphs that use rigid pointers, Rice uses fluid associations.

• Implicit Linking: Relationships are discovered based on usage and context, not just manual entry.
• Path Navigation: Agents can "follow" a thought process from one concept to another.
• Example: Identifying that "Project X" is associated with "Team Alpha" and "Budget 2024" without explicit foreign keys.

2. Working Memory (Scratchpad)

Just like humans have short-term memory for the task at hand, agents need a fast, ephemeral space for active context.

Key Characteristics

• High-Velocity: Optimized for the rapid read/write cycles of an active agent loop (latency < 1ms).
• Context-Aware: Keeps current conversation history and intermediate "thoughts" distinct from long-term knowledge.
• Time-Ordered: Essential for maintaining the sequence of a conversation or execution plan.

Session Management

• Isolation: Create separate memory sessions for different agent tasks (e.g., task-123, chat-user-456).
• Auto-Expiry: Set Time-To-Live (TTL) policies so old context doesn't clutter memory.

3. Access Control Lists (ACL)

In multi-user or multi-agent environments, data isolation is critical.

The Problem

Traditional vector databases often lack fine-grained security. Filtering results after search is slow and insecure.

The Rice Solution

• Zero-Latency Checks: Permissions are encoded into the search index itself.
• Document-Level Granularity: Control read, write, and delete permissions per user and per document.
• Bitmap Indexing: Uses efficient bitwise operations to filter millions of documents instantly.

4. Hyperdimensional Computing (HDC)

Under the hood, Rice uses advanced Hyperdimensional Computing techniques.

What is it?

A computing paradigm inspired by how the brain works, using large holographic vectors rather than numerical matrices.

Why it matters?

• Efficiency: Compresses complex data structures into fixed-size vectors, reducing memory footprint.
• Speed: Operations like "binding" (combining concepts) are single-pass mathematical operations.
• Reliability: Highly tolerant to hardware faults or bit-flips.