Skip to main content

Architecture Overview

System Overview

ContextKeeper anchors project memory behind a Sacred Layer that locks approved architectural plans while a RAG agent serves contextual answers to developers and AI assistants. The system is designed to keep day-to-day coding aligned with strategic intent.

Core Services

  • Sacred Layer Service — Immutable plan registry with dual-factor approval
  • RAG Agent — Clusters repository knowledge in ChromaDB and answers natural-language queries
  • Git Activity Tracker — Watches commits to detect drift from the Sacred Layer
  • CLI & MCP Tools — Shell tooling and Claude Code MCP endpoints for interactive access

Data Storage

  • ChromaDB stores per-project embeddings plus the Sacred Layer corpus
  • Project State Files persist objectives, decisions, and alignment metrics
  • Environment Variables (SACRED_APPROVAL_KEY, API tokens) gate access to write paths

Integration Points

  • Google Cloud GenAI supplies embeddings and responses when cloud inference is permitted
  • Local Security Mode falls back to on-device models for sensitive repositories
  • Claude Code MCP exposes Sacred-aware tooling directly inside the IDE

Security Model

  • Immutable plan hashes plus the Sacred Approval Key prevent unreviewed changes
  • Drift detection flags commits that contradict Sacred Layer directives
  • All secrets stay local; network calls are limited to configured AI providers
Architecture distilled from the upstream ContextKeeper repository and Sacred Layer specs.