Kenfold

•
72
•

Store, synchronize, and share persistent context across multiple AI agents over MCP using PostgreSQL and pgvector.

Kenfold is a self-hosted memory server engineered to unify context across disparate AI development tools such as Claude Code, OpenAI Codex, OpenCode, and ChatGPT over the Model Context Protocol (MCP). Built on PostgreSQL with pgvector, it treats memory as structured data with clear lifecycle types, strict agent provenance, and explicit version history to avoid silent overwrites. With intelligent session hooks, hybrid search combining full-text and vector embeddings, automated background memory extraction, and AST-based stale code detection, Kenfold ensures that decisions made in one environment seamlessly and reliably inform the next.

主な機能

0172 GitHub stars
02Hybrid search integrating PostgreSQL full-text search, dense vector embeddings, and cross-encoder reranking
03Session lifecycle hooks that automatically capture context, create summaries, and handle seamless cross-agent handoffs
04Tree-sitter-based code reference tracking to automatically detect and mark memories as stale when underlying code changes
05Poisoning-aware architecture with human-in-the-loop review dashboards for extracted memories, preferences, and consolidations
06Cross-agent memory sharing with strict provenance tracking via agent-specific API keys and OAuth 2.1

ユースケース

01Resuming development work across different AI programming assistants without losing project context or architectural decisions
02Tracking dynamic codebase knowledge and receiving automatic alerts when related functions or symbols are modified
03Capturing and consolidating project rules, preferences, and session histories across team workflows safely with human review