# Plugins
{/* AUTO-SYNCED SOURCE: this page lives in apps/app/src/modules/agent/docs/ and is mirrored into the docs app by `bun sync:module-docs`. Edit it in the module, not in apps/docs. */}
A plugin is something the assistant can do for you: work with onOffice, read properties, search the web, manage appointments. Under **Assistant > Plugins** (Assistent > Plugins, `/agent/plugins`) you see all plugins at a glance, with their state, their skills, and what leaves the platform in the process. There are two kinds: workspace plugins apply to the whole office, personal plugins belong to you alone.
## Where to find the plugins
In the assistant, the sidebar has the entry **Plugins** at the bottom. The overview shows a search box and a switch for **Alle** (all), **Für dich** (for you), and **Für den Arbeitsbereich** (for the workspace). Both are part of the address, so you can link to a filtered view. Every tile leads to the plugin's page.
Everyone who may use the assistant sees the overview, including members without management rights. Only people with the corresponding right can change anything there (see below). A member sees the same layout, just without switches.
## Workspace plugins
Workspace plugins apply to everyone in the office. They include the platform's capabilities (properties, knowledge vault, files, help and docs, web search, market and reference data, runs, tasks, communication) and every installed tool, such as DB-Leads, AI Homestaging, the credentials vault, or appointment booking. A tool that is not yet installed from the marketplace appears with the state **Nicht installiert** (not installed) and leads to the marketplace.
Every workspace plugin has exactly one switch for the whole and, below it, its **skills**. If the plugin is deactivated, the assistant no longer knows any of its skills. If it is active, every skill counts individually.
Only the owner activates a workspace plugin. Before activating, a dialog shows what happens, what leaves the platform, and how you stay in control.
Owners and admins deactivate a plugin. The assistant loses all of the plugin's skills immediately. The tool's stored data remains, it is just no longer used through the assistant.
Every tile carries its state as text: **Aktiv** (active) with the number of enabled skills, **Deaktiviert** (deactivated), **Nicht eingerichtet** (not set up, the integration behind it is still missing, for example onOffice under Settings > Integrations), **Störung** (disruption), **Nicht installiert** (not installed), or **Folgt** (coming).
## Skills
A skill is a single capability within a plugin, for example "search properties" or "set a follow-up". On a plugin's page the skills are grouped into **Lesen** (read) and **Schreiben** (write). Clicking a skill opens its dialog with description, examples, and the switch.
The assistant starts with full capability: all reading skills are enabled from the start. Some writing skills are locked by default because their effect is visible to the whole office, such as cancelling an appointment. You enable those deliberately.
The principle remains: **What is locked, the assistant does not execute.** It does know that the skill exists, though, and says so openly instead of pretending not to know it. If you ask for a web search and web search is locked, that is exactly the answer you get, plus a pointer to where it is enabled.
## The connector comes first
Once you have connected your personal onOffice AI Connector, the assistant works in onOffice for you exclusively through its tools, with your CRM permissions. The reading skills of the onOffice workspace plugin step back for you; whoever has no connector keeps working with them. Activity log entries, appointments and tasks in onOffice are created only through the connector, and the entry carries your name. Without a connector the assistant says so openly and shows you the way to connect instead of filing the note somewhere else.
In the chat the assistant's steps read in plain language, for example "Lege Maklerbuch-Eintrag in onOffice an", and below the answer you see which plugins took part, for example "Mit onOffice AI Connector beantwortet (2 Abfragen, 1 Änderung)". Only successful calls are counted.
Enabled does not mean unasked: even with full permissions, each person confirms change actions individually in the chat via the confirmation cards. The skill switch decides whether the assistant can do something at all. The card in the chat decides whether it does it now.
Above the skill list there are two bulk switches: **Alle freigeben** (enable all) and **Alle sperren** (lock all). When enabling a writing skill, the page asks once.
## Personal plugins
A personal plugin belongs to exactly one person. It applies only to you, it appears only in your overview, and only you can set it up or disconnect it. Owners and admins see in the workspace plugin onOffice who in the office is connected and since when; what the assistant does on your behalf, they do not see there.
The first personal plugin is the **onOffice AI Connector**. It connects the assistant to your own onOffice account so that it works with your permissions: it sees exactly what you may see in onOffice, nothing more. This lets the assistant, for example, read your activities, addresses, and tasks and, after confirmation, create entries.
The workspace plugin **onOffice** is active, and an owner or admin has set the approval "Mitarbeitende dürfen den onOffice AI Connector für sich einrichten" (staff may set up the onOffice AI Connector for themselves) there. As long as the approval is missing, the personal plugin is invisible to members.
On the plugin's page, **Verbinden** (connect) first shows a dialog with what happens. Then you are taken to onOffice and sign in there. onOffice asks whether the assistant may act on your behalf. After your yes you are back on the plugin page and the state reads **Verbunden** (connected).
Only the access authorisation, sealed on the platform. The platform never sees your onOffice password.
**Trennen** (disconnect) revokes the access authorisation immediately. The assistant can no longer act on your behalf afterwards. You can reconnect at any time.
If the authorisation expires or onOffice declines, the page shows **Störung** (disruption) and offers **Neu verbinden** (reconnect). The assistant also tells you in the chat, with a link to the plugin page.
## Approval by the administration
On the page of the workspace plugin **onOffice**, owners and admins see the switch **Mitarbeitende dürfen den onOffice AI Connector für sich einrichten** (staff may set up the onOffice AI Connector for themselves). It is off by default. Below it, the page lists the people in the office who are currently connected, with the date of the connection and a note if a connection is disrupted.
If you turn the approval off while connections exist, a dialog asks what should happen to them:
The existing connections stay stored, but the assistant no longer uses them. If you turn the approval back on later, they continue without reconnecting.
All connections are revoked. After a new approval, every person has to reconnect themselves.
Owners and admins always see the personal plugin, even without approval, so they can review it and set it up for themselves.
## Who may do what
Sees everything, activates and deactivates workspace plugins, toggles skills, sets the approval for the personal connector, and connects themselves.
Like the owner, with one exception: **activating** a workspace plugin is owner-only. Deactivating, toggling skills, and setting the approval are open to admins.
Sees the overview and every plugin page without switches. Sets up personal plugins for themselves once they are approved.
Sees the overview and the plugin pages, changes nothing, and does not set up personal plugins.
The rights belong to the **Assistent** (assistant) family in the permission management under Settings > Team and can be granted individually in custom roles: read plugins, manage plugins, connect personal plugins, activate plugins.
## What leaves the platform
Every plugin page has the section **Was hinausgeht** (what goes out). It states which data leaves the plugin and where to: to the AI provider in the EU, to onOffice, to the web search. A plugin without this information shows the gap visibly instead of hiding it. You find the same information in the processing records under Settings > Privacy.
## What came before
Until 5 September 2026 there was the **permissions console** under Settings > Assistant and the **Verbindungen** (connections) section under Account > Assistant. Both have merged into the plugins: every group of the console is now a workspace plugin, every permission a skill, and the onOffice AI Connector is a personal plugin. Your previous permissions and connections were carried over. The console's old address redirects to the plugins.
## Further reading
What the individual skills do is covered in the [capabilities reference](/tools/agent/faehigkeiten). The remaining assistant settings are in [Settings](/tools/agent/einstellungen). Back to the [overview](/tools/agent).