Conectar

Krusch Context

kruschdevbykruschdev
•
83
•
Gestión de Bases de Datos
Herramientas para Desarrolladores
Productividad y Flujo de Trabajo

Provide persistent, cross-session working memory and project rule governance for AI coding agents using local SQLite or PostgreSQL storage.

Krusch Context is a high-performance long-term memory layer engineered for AI coding agents in Cursor, Claude Code, Windsurf, and Antigravity. By leveraging a lightweight, five-verb architecture using built-in Node.js SQLite (or optional PostgreSQL with pgvector), it enables agents to store, retrieve, revise, and audit project decisions, architectural invariants, and bug fixes without token waste or continuous external LLM costs. It features near-duplicate detection, span-grounded provenance to heal code drifts, and pre-commit diff auditing to keep AI agents aligned with core project standards over long-term workflows.

Características Principales

015 canonical MCP verbs (retrieve, remember, revise, nudge, health) to minimize token overhead
02Closed taxonomy enforcement (decision, invariant, bug, lesson, blocker) with near-duplicate detection
03Span-grounded provenance with SHA-256 anchoring and automatic drift self-healing
04Pre-commit diff auditing via the nudge tool to prevent invariant violations
05Zero-GPU offline support with keyword extraction, alongside optional local Ollama or cloud embeddings
0682 GitHub stars

Casos de Uso

01Tracking bug resolutions and diagnostic findings so agents do not repeat past debugging loops
02Maintaining architectural decisions and non-negotiable code invariants across multiple agent sessions
03Auditing code diffs before commits to prevent agents from reintroducing known bugs or anti-patterns