AI assistant
The Assistant page on the hub lets you ask questions and give tasks in plain language. The assistant uses tools to read real data from your sites and, if you allow it, to write drafts.
Set an OpenAI-compatible provider on the hub server:
WUPPY_AI_BASE_URL=https://api.example.com/v1WUPPY_AI_API_KEY=...WUPPY_AI_MODELS=deepseek-v4.1-flash # comma-separated; the first one is the defaultAny OpenAI-compatible endpoint works (OpenAI, OpenRouter, Ollama, EVREN…). The model must support tool calls. The key stays on the server and is never sent to the browser.
Agents
Section titled “Agents”| Agent | What it does |
|---|---|
| Agency assistant | Any task across the sites you can reach |
| Site care | Health checks, connection problems, security scans, recent activity |
| Client reporter | Writes client reports from real data and can email them to a site member |
| Content editor | Drafts and edits content in the client’s voice |
Choose an agent and, optionally, a site. Answers stream in as they are written, and each tool the assistant runs is shown with its result.
Changes need your permission
Section titled “Changes need your permission”Writing tools are off by default. Turn on Allow changes for a message to let the assistant create or edit content. New content is always saved as a draft; the Changes to review panel links to each one so a person can check and publish it.
The assistant acts with your permissions: every tool call is a normal API request checked like any other. Tool output is treated as data, never as instructions.
Context notes
Section titled “Context notes”On Sites, each site has two notes that the assistant reads first:
- Agency playbook: how your agency works (tone, rules, what to check).
- Client brief: the client’s brand voice, audience and wishes.
Only users with sites:manage can read or edit them.
Saved chats
Section titled “Saved chats”Chats are saved per user; only you can list, open or delete yours. Start a new chat or reopen an old one from the list. Each user keeps up to 200 chats.
Limits
Section titled “Limits”20 messages per minute per user, at most 8 tool steps per answer, and a provider timeout.