Hila Shmuel: Cabinet release
- Ex‑Apple engineering manager Hila Shmuel announced Cabinet, an LLM+knowledge‑base tool with cloud agents and Kanban workflows. - The project has already reached roughly 1.5k stars on GitHub in early public visibility. - Cabinet represents a practical, repo‑level example of agent workflows and knowledge integration for developer productivity tooling. (x.com)
A new open-source project called Cabinet is gaining traction by turning a folder of markdown files into an AI knowledge base with built-in agents. (github.com) The repository, published by Hila Shmuel, showed about 1.4k GitHub stars and 133 forks when checked on April 19, 2026. The README describes Shmuel as a former engineering manager at Apple and says Cabinet is being built in public. (github.com) Cabinet stores work as files on disk instead of in a hosted database, and its site says it is free, open source, self-hosted, and designed to avoid vendor lock-in. The project’s pitch is a knowledge base plus “AI team,” with agents, workspaces, and memory tied to local files. (runcabinet.com, github.com) That design puts Cabinet in a fast-moving part of developer tooling: software that gives language models persistent context instead of a blank chat window each session. Cabinet’s README frames the problem as scattered documentation, forgotten project history, and repeated copy-paste between tools. (github.com) The project also packages agent workflows in a form developers already recognize: folders, Git history, release tags, and desktop builds. GitHub’s own product team has been pushing “agentic workflows” inside repositories, including automations for documentation, code quality, and triage. (github.com, github.blog) Cabinet’s recent releases show how quickly the project is moving. The release log from two to three days before April 19 lists a signed macOS Electron app, a command-line installer, GitHub Actions-based publishing, and 24 agent personas available on a fresh clone. (github.com) The onboarding flow is aimed at small teams and solo builders rather than large enterprise deployments. A mirrored README says users can start with `npx create-cabinet@latest`, answer five onboarding questions, and generate a custom AI team in about two minutes. (github.com) Cabinet’s public code also shows the tradeoffs of this approach. Keeping everything in local files can make versioning and portability easier, but the project still depends on model providers, runtime setup, and agent orchestration to make those files useful in practice. (github.com, github.com) For now, Cabinet looks less like a polished enterprise suite than a live example of how AI agents, local knowledge, and developer workflows are being fused into one repo. Its early GitHub traction suggests there is already an audience for that model. (github.com)
