CentR Icon CentR v0.1.0
GitHub
Introduction
Getting StartedArchitecture
Core Intelligence
Project MemoryGlobal LearningOptional Local Brain
Tools & Integrations
Model Context Protocol (MCP)CLI Command Reference
Engineering & Standards
Security & PrivacyBenchmarks & ValidationContributing Guide

CentR Architecture Deep Dive

CentR is architected on a single inviolable principle: Deterministic Core is the authority; the optional Local Brain is an advisor.

Architecture Diagram

The Two-Tier Subsystem

1. Deterministic Core (The Authority)

  • AST Indexer: Extracts qualified function signatures, classes, interfaces, and exported symbols.
  • SQLite 3 + WAL Mode: Fast concurrent in-process database storage.
  • FTS5 BM25 Engine: Sub-2ms full-text ranking across symbols, file paths, and project memory.
  • Token Budgeter: Greedy relevance packing to fit within context limits (e.g. 4000 tokens).
  • Project Memory: Project-isolated institutional memory.
  • Global Learning: Cross-project evidence-backed knowledge.

2. Optional Local Brain (The Advisor)

  • Powered by local Small Language Models (0.5B–7B parameters via Ollama or custom local providers).
  • Semantic re-ranking, task classification, and failure diagnosis.
  • Hallucination Guard: Strictly prunes any invented candidate IDs; only Core candidates can be returned.
  • Deterministic Fallback: If the local SLM is absent, slow, or times out, CentR automatically falls back to Core heuristics.