cabinet: Build a Self-Hosted AI Knowledge Base

Overview

Cabinet is a self-hosted workspace that pairs a Markdown knowledge base with AI agents and operational tools. Its files live in a directory on disk, with Git-backed history, so project context can accumulate across agent sessions without being confined to one AI provider’s storage.

It is intended for individuals and small teams who want to organize startup work and delegate repeatable tasks to AI agents. Compared with a conventional notes app, Cabinet adds agent orchestration, scheduled jobs, and embedded HTML apps. It makes the most sense when you are comfortable running a local web application and managing AI CLI providers.

Key Features

• Stores knowledge as Markdown files on disk, without a database.
• Tracks edits with Git-backed history, diffs, and page restoration.
• Provides AI agents with workspaces, goals, skills, and scheduled jobs.
• Supports local Claude Code and Codex CLI adapters for agent tasks and jobs.
• Organizes goals into missions and tasks, with Kanban boards and internal chat.
• Renders HTML apps placed in knowledge-base folders and includes a web terminal.
• Offers rich-text Markdown editing, full-text search, and PDF and CSV viewers.

Use Cases

• Founders can keep company plans, research, and decisions in one local knowledge base and reuse that context across AI sessions.
• Small teams can assign recurring research, reporting, or content work to agents and review the resulting files in a shared workspace.
• Developers can prototype lightweight tools by placing HTML files in the knowledge base, with changes tracked through Git.
• People moving between AI providers can retain portable project context instead of relying on provider-specific memory.

Getting Started

npx create-cabinet@latest
cd cabinet
npm run dev:all

Open http://localhost:4000. See the repository README for setup, configuration, and CLI details.

Alternatives

• TencentDB-Agent-Memory: TencentDB-Agent-Memory focuses on shared team context for agents, while Cabinet centers on a local Markdown knowledge base with automation.
• Cognithor: Cognithor is a local autonomous-agent platform with broad model and channel support, while Cabinet organizes persistent context in Markdown files.
• taOS: taOS provides a self-hosted agent desktop for chats, files, and memory, while Cabinet emphasizes a Markdown knowledge base and scheduled workflows.

Considerations

• Requires Node.js 22 or later according to the README, plus at least one supported AI CLI provider: Claude Code or Codex.
• Cabinet runs locally and self-hosts its application, so users are responsible for the runtime environment and access controls. A password gate is optional and disabled when KB_PASSWORD is empty.
• The project sends anonymous usage telemetry by default. It can be disabled through an environment variable or the settings UI.
• Agent capabilities depend on the configured CLI provider; the project does not present itself as a hosted AI service.
• The repository describes an actively evolving product and asks prospective contributors to coordinate with the maintainer before opening a pull request.

Repository: cabinetai/cabinet