Rementum
It gives AI agents a self-hosted shared memory infrastructure: versioned Markdown knowledge, encrypted article bodies, routing and semantic retrieval, and conflict-safe staged writes coordinated over MCP.
README
Rementum

Rementum is an open-source, self-hosted shared brain for AI agents. It gives Claude, ChatGPT, Codex, Cursor, and any remote MCP client one versioned, auditable body of knowledge.
The product is deliberately agent-first:
- Knowledge lives in linked Markdown articles with a compact routing index.
- Every canonical change is staged, versioned, attributed, and conflict checked.
- Article bodies are encrypted with a per-brain key.
- Rementum creates a compact routing summary locally or with an optional OpenAI-compatible AI provider.
- Search combines routing metadata, PostgreSQL full-text search, and local multilingual embeddings.
- Agents coordinate work through leased tasks and write maintenance proposals back through the same staged protocol.
- The whole brain exports as portable Markdown.
Status
This repository is under active development toward a production beta. The REST and MCP contracts are versioned, but backward compatibility is not promised until 1.0.
Self-host
Point a domain at a Linux host with Docker Compose, open ports 80 and 443, then run:
git clone https://github.com/yibudak/rementum.git
cd rementum
./scripts/install.sh
The installer generates the instance secrets, builds the stack, runs migrations, waits for health checks, and creates the first owner. Caddy provisions HTTPS. See the installation guide for requirements and recovery steps.
After configuring encrypted backups, update an installed instance with:
./scripts/update.sh
It backs up the instance, fast-forwards the source, runs migrations, rebuilds the services, and waits for their health checks. See the operations guide for backup setup and recovery.
Documentation
The MkDocs site covers configuration, backups, upgrades, security, and agent connections. Build it with:
python3 -m pip install -r docs/requirements.txt
mkdocs build --strict
For development setup and checks, read docs/development.md.
Security boundary
Article bodies and version bodies are encrypted at rest. By default, Rementum creates routing summaries locally and does not send staged bodies to an external LLM. If you enable an OpenAI-compatible provider, Rementum sends it the complete resulting article body in plaintext before encryption. Routing metadata and embeddings remain searchable and must be treated as sensitive derived data. The master key is not stored in the database or included in backups.
See SECURITY.md and the security checklist before storing private knowledge.
Backup
Set REMENTUM_BACKUP_AGE_RECIPIENT to an age public recipient and run:
docker compose --profile backup run --rm backup
The encrypted archive contains PostgreSQL, local blobs, and a versioned manifest. It never contains
REMENTUM_MASTER_KEY; escrow that key separately. Read the operations guide
before you test a restore.
License
AGPL-3.0-only. See LICENSE.
推荐服务器
Baidu Map
百度地图核心API现已全面兼容MCP协议,是国内首家兼容MCP协议的地图服务商。
Playwright MCP Server
一个模型上下文协议服务器,它使大型语言模型能够通过结构化的可访问性快照与网页进行交互,而无需视觉模型或屏幕截图。
Magic Component Platform (MCP)
一个由人工智能驱动的工具,可以从自然语言描述生成现代化的用户界面组件,并与流行的集成开发环境(IDE)集成,从而简化用户界面开发流程。
Audiense Insights MCP Server
通过模型上下文协议启用与 Audiense Insights 账户的交互,从而促进营销洞察和受众数据的提取和分析,包括人口统计信息、行为和影响者互动。
VeyraX
一个单一的 MCP 工具,连接你所有喜爱的工具:Gmail、日历以及其他 40 多个工具。
graphlit-mcp-server
模型上下文协议 (MCP) 服务器实现了 MCP 客户端与 Graphlit 服务之间的集成。 除了网络爬取之外,还可以将任何内容(从 Slack 到 Gmail 再到播客订阅源)导入到 Graphlit 项目中,然后从 MCP 客户端检索相关内容。
Kagi MCP Server
一个 MCP 服务器,集成了 Kagi 搜索功能和 Claude AI,使 Claude 能够在回答需要最新信息的问题时执行实时网络搜索。
e2b-mcp-server
使用 MCP 通过 e2b 运行代码。
Neon MCP Server
用于与 Neon 管理 API 和数据库交互的 MCP 服务器
Exa MCP Server
模型上下文协议(MCP)服务器允许像 Claude 这样的 AI 助手使用 Exa AI 搜索 API 进行网络搜索。这种设置允许 AI 模型以安全和受控的方式获取实时的网络信息。