Sibyl
Store and share persistent cross-agent memory and project context across AI coding tools using a self-hosted knowledge graph.
Sibyl is a self-hostable cross-agent memory platform that gives developers and AI coding assistants a unified, persistent knowledge graph. Instead of losing context between sessions or locking memory into proprietary vendor silos, Sibyl allows tools like Claude Code, Codex, Cursor, and custom autonomous agents to share learned decisions, gotchas, error patterns, and tasks. Built on SurrealDB and accessible via a CLI, web dashboard, and Model Context Protocol (MCP) server, it implements a continuous context loop (context → act → remember → reflect) that compounds project intelligence over time.
主な機能
0169 GitHub stars
02Self-hostable, multi-tenant architecture powered by SurrealDB with source-grounded synthesis and document crawling
03Durable cross-agent memory loop (context → act → remember → reflect) shared across tools like Claude Code, Cursor, and custom agents
04Semantic search and graph exploration for contextual recall across decisions, conventions, and architectural rules
05Flexible access model supporting native CLI scripting, agent skills, web dashboard, and an MCP interface
06Integrated task workflow management spanning backlog, active execution, and task-derived learnings
ユースケース
01Preserving architectural decisions, debugging patterns, and conventions across different AI coding tools and sessions
02Compiling compact, project-scoped context packs to bootstrap coding agents before tackling complex refactors or builds
03Tracking multi-step development tasks and automatically capturing persistent insights upon completion