Vellum has launched Plugin Hub, an open catalog of installable plugins for its local-first personal AI assistant. This release distinguishes between skills and plugins: while a skill handles a single task such as drafting one email, a plugin provides the assistant with a standing capability it can execute independently. This includes tasks such as connecting to Gmail and reading your inbox, managing a calendar, tracking expenses, syncing notes, or monitoring code repositories. Each plugin is positioned as a complete function the assistant owns, rather than a one-off prompt.
Plugins are developed as TypeScript packages with lifecycle hooks, differing from skills, which are instruction bundles. The manifest utilizes the vellumai plugin-api and exposes hooks for init, user-prompt-submit, post-tool-use, post-model-call, and on Shutdown, with state persisted to a local JSONL file. This design is intended to be more accessible than authoring a full MCP server, allowing more people than just infrastructure engineers to build one. For those who prefer not to write code, Vellum offers a plugin builder skill that generates a plugin from a plain-language description.
A key feature of this setup is credential handling. When a plugin requires an API key or an OAuth token to access email, banking, or a code repository, that credential remains on the machine and is not sent to the model. This ensures the assistant can be trusted with sensitive integrations. Plugins are open-source and portable, allowing them to be shared among users rather than tied to a single vendor. A single plugin can also operate across different platforms: installed on macOS, the same plugin and assistant can run through Slack and Telegram with shared memory, making a capability built once available wherever the assistant is accessed.
The launch catalog includes a variety of use cases:
Marketing Expert: Bundles positioning, launches, content, brand voice, and reporting into one plugin.AI Hero Engineer Kit: Packages engineering workflows around plans, PRDs, vertical-slice issues, and test-driven loops.Writing Coach: Runs daily prompts that escalate with a streak and critiques drafts.Fitness Companion: Logs meals, workouts, and weight, pulls nutrition data from OpenFoodFacts, and calculates macro targets.
Plugins can be installed from the catalog or directly from a GitHub URL, and authors can publish their own to the same catalog.
Start testing Vellum!
Take me there! Vellum is a local-first personal AI assistant that operates as a native macOS app or as a self-hosted service, with an open-source runtime and support for multiple model providers, including Anthropic, OpenAI, Google, and local models via Ollama. The assistant is built around persistent memory, its own identity, and permission-gated action on the user's machine. Plugin Hub extends this foundation into a shared ecosystem, transitioning the assistant from handling one-off tasks to managing ongoing parts of a workflow. The company describes the catalog as the layer that transforms the assistant from a product into a platform.