System Architecture Reference

Home · Architecture · Reasoning · Theory · Syntax · APIs · Wiki · Specs · Research Layered design for neuro-symbolic reasoning

AGISystem2 follows a layered architecture separating core HDC operations, runtime management, and high-level reasoning. This design enables modularity, testability, and clear responsibility boundaries.

System Overview Diagram

AGISystem2 Architecture Layers
API Layer learn() query() prove() summarize() elaborate() Processing Layer Parser Executor Query Engine Proof Engine Decoder Runtime Layer Session Scope Theory Registry Knowledge Base Vocabulary Core HDC Layer Vector Operations Position ASCII Stamp Constants

Layer Descriptions

API Layer

Public interface for developers. Provides high-level methods for learning facts, querying knowledge, proving goals, and generating explanations. Designed for ease of use and integration.

  • learn(dsl) - Add facts and definitions
  • query(dsl) - Find answers with holes
  • prove(goal) - Build proof trees
  • summarize() - Concise output
  • elaborate() - Detailed narratives

Processing Layer

Core logic for transforming DSL input into vector operations and back. Handles parsing, execution, query resolution, proof search, and result decoding.

  • Parser: DSL text to AST
  • Executor: AST to vector operations
  • Query Engine: Hole-filling search
  • Proof Engine: Backward chaining
  • Decoder: Vector to structure

Runtime Layer

State management for sessions, variables, theories, and knowledge. Maintains isolation between sessions and manages the lifecycle of reasoning contexts.

  • Session: Isolated reasoning context
  • Scope: Variable hierarchy
  • Theory Registry: Loaded theories
  • Knowledge Base: Bundled facts
  • Vocabulary: Known atoms

Core HDC Layer

Fundamental hyperdimensional computing operations. Strategy-agnostic interface for binding, bundling, similarity, and deterministic initialization. Supports multiple HDC strategies.

  • HDC Facade: Strategy-agnostic API
  • Operations: bind(), bundle(), similarity()
  • Position: Pos1-Pos20 vectors
  • Strategies: dense-binary, sparse-polynomial, metric-affine, metric-affine-elastic (EMA), exact (lossless)
  • Constants: Geometry, thresholds

Data Flow

Query Execution Pipeline
DSL Text @q loves ?who Mary Parse AST nodes Build Partial Skip holes Unbind KB UNBIND Decode/Rank Candidates Result ?who = John

Design Principles

Principle Description Benefit
Layered Separation Each layer has clear responsibilities and interfaces Testability, maintainability
Determinism First All operations produce identical results for identical inputs Reproducibility, debugging
Minimal Operations Only Bind (XOR) and Bundle (Majority) at core Mathematical elegance, provability
Session Isolation Sessions are independent reasoning contexts Multi-tenancy, parallelism
Theory Composability Theories can be loaded and stacked Reusability, domain separation
Confidence Transparency All results include confidence scores Trust calibration

Related Documentation

AGISystem2 architecture is designed for deterministic, explainable reasoning with clear separation of concerns.