All topics

AI assistant and agents (MCP)

The assistant and AI agents are optional. Nothing is sent to a provider until an administrator enables AI and accepts the data notice.

Available options depend on your version, permissions and hardware. For missing actions, see Security and Diagnostics.

Step by step

  1. In System > AI, choose a provider (Claude, OpenAI or an Ollama on your network), enter its key or address and test the connection.
  2. Turn on the assistant and open it with the button at the bottom right; ask about your data or attach photos, PDFs or text files.
  3. For external agents, create a token in AI agents (MCP), copy the endpoint and configuration, and add them to your MCP client.

Options explained

Provider
Claude and OpenAI process the conversation in their cloud; Ollama keeps it on your network.
Model id
use the provider's model identifier, not its display name.
Memory
the assistant remembers earlier conversations; you can review or erase what it keeps.
Proposed changes
only if an administrator allows it; you apply each change yourself with a button.
Attachments
up to 4 files per message; photos are reduced before sending and files are deleted with the conversation.
MCP tokens
give each agent its own token and revoke it when you no longer use it.

Troubleshooting

If the connection test fails, read the provider's message on screen, check the key, workspace and model id, and confirm the device can reach the provider.

Common problems
Contribute