Getting started

onplate is an app that puts the “skills” of your AI tools in order. Scanning configs, estimating context and parsing local logs all happen on your own computer; no sign-up and no email. The network is used when you work with Git sources, marketplaces, updates or remote resources yourself.

Skills and MCP in plain words

If you use AI assistants (Claude Code, Codex, GLM, OpenCode), you can extend them with:

• Skills — text instructions that teach an AI to do something specific: “format reports from this template”, “check texts for these mistakes”. Technically it is a SKILL.md file, but you never have to open it by hand — onplate shows everything in the interface.
• MCP servers — small extensions that give an AI extra abilities: look into a database, call an external service, work with files.

The trouble is that every AI tool has its own folders and formats, and over time it stops being obvious what is installed, how much context it takes and what you actually use. onplate first scans your existing configuration, then shows usage, risks and outdated installs. Portable skills can be stored once and installed into every AI that supports them.

Install

The app is available for macOS, Linux and Windows. Download a build on the home page.

macOS

1. Download the onplate-macos-arm64.zip archive.
2. Double-click the archive — it unpacks into the onplate app.
3. Drag onplate into your Applications folder.
4. On the first launch macOS may warn you that the app was downloaded from the internet. Right-click the icon → Open → confirm Open again. You only have to do this once.

Linux

1. Download the onplate-linux-amd64.tar.gz archive.
2. Unpack it into a folder of your choice, for example:

   tar -xzf onplate-linux-amd64.tar.gz -C ~/apps/onplate
   

3. Install the WebKitGTK 4.1 system library if it is missing — onplate uses it to draw the interface:

   # Ubuntu 22.04+ / Debian 12+
   sudo apt install libwebkit2gtk-4.1-0
   
   # Fedora
   sudo dnf install webkit2gtk4.1
   
   # Arch
   sudo pacman -S webkit2gtk-4.1
   

4. Run the onplate executable from the unpacked folder — double-click it in the file manager or start it from the terminal:

   ~/apps/onplate/onplate
   

If the launch fails with error while loading shared libraries: libwebkit2gtk-4.1.so.0, install the package for your distribution: libwebkit2gtk-4.1-0 on Ubuntu/Debian, webkit2gtk4.1 on Fedora, webkit2gtk-4.1 on Arch (the ready commands are in step 3).

The minimum distributions where WebKitGTK 4.1 is in the standard repositories: Ubuntu 22.04, Debian 12, Fedora 36. Older releases do not have the library — you will need to update the system.

Windows

1. Download the onplate_<version>_windows_amd64_setup.exe installer and run it.
2. On the first launch Windows SmartScreen may show a blue “Windows protected your PC” window — the build is not signed with a publisher certificate yet. Press More info → Run anyway.

If running installers is blocked by your organization’s policy, take the portable archive onplate_<version>_windows_amd64.zip: unpack it anywhere and run onplate.exe — no install and no administrator rights needed.

First launch

On the first launch the storage is empty, and the dashboard greets you with three buttons — three ways to fill onplate up:

1. Import a package — add a set of skills from a folder on disk or from a Git repository. Use this if somebody sent you a link to a collection of skills. See Packages.
2. Connect a repository — track skills from a Git source and get updates automatically.
3. Scan the system — find the skills already installed directly in your AI tools and take them under management. See Sync.

If none of that is you yet — just open the Marketplace and find your first skill by searching.

How the interface is laid out

The sidebar is on the left. The Dashboard sits on top, then three groups (this documentation follows the same structure):

Group	Sections	What for
Manage	Skill storage, Trash, MCP servers, Subagents, Bindings, Editor	What you have and how it is installed
Discover	Marketplace, Repositories, Sync	Where new skills, MCP servers and subagents come from
Control	Security, Conflicts, Activity log, Team	Safety, contradictions, history and the shared team set

At the bottom of the menu are Settings (theme, language, behavior and install paths) and the theme switch.

Orbi — help for the current screen

Orbi lives at the bottom of the sidebar and works as a local, contextual helper. It is not a general AI chat: instead of free-form conversation it explains only what onplate already knows about the current screen and the local state.

Open Orbi to:

• understand what the current section is for and get a recommended first step;
• see what deserves attention before you act;
• unfold a short breakdown of how the screen is put together;
• check the last confirmed action and jump to the related bindings, storage, packages, trash or activity log;
• look through up to ten recent explanations of actions, stored locally.

If an operation finished without a confirmed final state, Orbi honestly marks the result as unconfirmed and suggests refreshing the matching screen first. That matters: an error message does not always mean nothing changed on disk.

Orbi closes with the cross or the Esc key. To remove it from the menu entirely, press Hide Orbi. Undo is available right away; later you can bring the helper back with ⌘/Ctrl+K or in Settings.

Orbi uses only the screen context the app already knows. File contents and secrets are never sent anywhere.

What next

Walk through the Quick start — in five minutes you will install your first skill into your AI tool and see the whole onplate cycle.