Cabinet, an open-source platform described as "the AI-first startup OS," has gained 393 GitHub stars since launching on April 3, 2026. Created by developer @hilash, Cabinet addresses a critical limitation of current AI assistants: the loss of context between sessions that requires users to repeatedly provide background information.
Persistent Knowledge Base Enables AI Agent Memory
Cabinet maintains organized knowledge stored as markdown files on disk, eliminating the need for traditional databases. The platform uses Git for version control, providing full diff viewing and version history for all content. Files can be edited through both WYSIWYG and markdown interfaces powered by Tiptap, with full-text search across all pages.
This file-based architecture ensures data ownership and transparency while allowing AI agents to access and build upon information over time. Unlike conversational AI assistants that start fresh with each session, Cabinet creates persistent memory that agents can reference across all interactions.
Twenty Pre-Built Agent Templates Cover Cross-Functional Roles
The platform includes 20 agent templates spanning different organizational functions: CEO, Content Marketer, DevOps Engineer, Product Manager, Sales Rep, and Customer Support, among others. These agents can communicate through internal team channels, operate on cron-based schedules for recurring tasks, and maintain organized workspaces within the knowledge base.
Agents access the shared knowledge repository to perform their functions, creating an AI-first workflow where automated team members are primary consumers of organizational information. This differs fundamentally from human-centric tools like Notion or Confluence.
Technical Stack Emphasizes Self-Hosting and Data Control
Cabinet is built with Next.js 16, TypeScript, Tailwind CSS, and shadcn/ui components. The architecture includes a Next.js frontend, a daemon server managing job scheduling and agent execution, and file-based storage instead of databases.
Unique features include embedded HTML applications (users can drop index.html files to create custom apps), a web-based terminal with Claude Code integration, built-in PDF and CSV viewers, and Kanban boards for task tracking. The platform follows a "BYOAI" (bring your own AI) philosophy, avoiding cloud vendor lock-in.
Use Cases Range from Solo Developers to Startup Teams
Cabinet targets startups maintaining institutional knowledge, solo developers managing multiple projects, teams coordinating AI agents for different functions, and companies seeking AI assistance without cloud dependencies. The self-hosted approach provides complete control over sensitive information while maintaining the flexibility to integrate various AI models.
The project emphasizes data ownership, transparency through git history, AI flexibility, simplicity, and security as core design principles.
Key Takeaways
- Cabinet gained 393 GitHub stars as an AI-first knowledge base storing everything as markdown files on disk
- The platform includes 20 pre-built agent templates covering roles from CEO to DevOps Engineer
- Git-backed version control provides full history and transparency for all knowledge base changes
- Built with Next.js 16 and TypeScript, Cabinet emphasizes self-hosting and data ownership over cloud dependencies
- The "BYOAI" approach allows users to integrate any AI model without vendor lock-in