An AI plugin extends your AI assistant with new capabilities — hooks, tools, and skills that run locally on your machine. Each plugin is a self-contained GitHub repo you install with a single command. Vellum plugins can compress the assistant's output, log your workouts, manage your calendar, and much more.

Three ways, all one step. Browse and install from the Plugins tab inside the Vellum app, send your assistant the plugin's GitHub link or its page on this hub and ask it to install, or run the CLI command from any plugin's page: assistant plugins install <name>. No build step, no config file to edit.

You don't have to do anything — just talk to your assistant like normal. Each plugin describes what it can do, and the assistant reads those descriptions and picks the right tool or skill when your request calls for it. Ask to search your Drive and a Drive plugin answers; ask for a workout log and a fitness plugin steps in. No commands to memorize, no menus to dig through.

Claude plugins and Claude code plugins run inside Anthropic's closed ecosystem — your plugin only works within Claude, and you're bound by Anthropic's terms. Vellum plugins are open source and portable: the source lives on GitHub, installs are pinned to a reviewed commit, and your credentials never reach the model. The same plugin runs across macOS, Slack, and Telegram with shared memory.

A skill is a single packaged capability. A plugin is broader: it can bundle skills together with hooks and tools that wire more deeply into the assistant. Skills ship inside Vellum; plugins are community and partner extensions distributed through this directory and installed from their own repos.

Yes. Every plugin in the directory is open source and free to install. Some plugins call third-party APIs that may have their own costs, but the plugin itself never costs anything.

Plugins in this directory are reviewed by the Vellum team, install from a pinned commit, and run inside Vellum's sandbox with per-actor permissions. Your credentials live in a separate process and never reach the model. You can also read every line of source on GitHub before installing.

In its own GitHub repo. Each plugin page links straight to the source at the exact commit it installs from, so you can inspect the scripts it runs and its full history before and after you install.

You stay on the version you installed until you choose to update, because installs are pinned to a specific commit. When you update, you can see exactly what changed in the repo first.

Absolutely. A plugin is a repo with a package.json manifest plus the hooks, tools, and skills it contributes. The simple-memory plugin is the reference implementation and exercises every wired surface — start there. Read the docs.

Only what you allow. Plugins run under Vellum's permission system: high-risk actions require your explicit confirmation, network access goes through an outbound proxy, and secrets are collected through a secure prompt that never exposes them to the model.

Ask your assistant to uninstall it, or run assistant plugins uninstall <name>. Plugins are self-contained, so removing one leaves the rest of your setup untouched.