The AI assistant and its tools
The dashboard's assistant that answers and acts on the agency's properties, inquiries and viewings, the tools behind it, and the MCP server that offers the same tools to coding agents.
For the Full Stack package
What the assistant does
The assistant, at /en/dashboard/ai-assistant, is a chat over the agency's own data. It answers from the properties, inquiries and viewings in the database, draws charts of inquiries and viewings, and acts on them when asked: draft a property, move an inquiry along, add a note, confirm or cancel a viewing, change an agency setting.
- A property can be picked as the conversation's context, and past conversations are kept per admin, so the history survives a reload. Each conversation is named after its first question, and a reopened one shows every reply under the question it answers.
- Answers stream as they are written, with the model's reasoning shown beside them on Gemini and DeepSeek models (OpenAI's answer without it), and a failed answer offers a Try again button.
- It can draft a listing's description in English and Arabic from the property's real facts.
What it needs and who can use it
- A key on the API for Google Gemini, OpenAI or DeepSeek, your own account with that provider. Any one is enough, and with several the picker offers the models of each. Without one the page says "The assistant needs an AI provider key" and lists the three keys it accepts, and nothing is billed.
- Each answer is billed by the chosen model's provider to your key for it, at that provider's price for the model picked in the assistant's picker.
- The staging studio is separate: it needs a Google Gemini key whichever provider the assistant uses.
- The
ai_chat.usepermission. Of the seeded roles only the Super Admin holds it; the Agent role does not see the assistant. - Nothing beyond the admin's own role: the assistant reads and changes only what the admin's other permissions allow, and checks again before every action.
The MCP server
The API also offers the assistant's tools over the Model Context Protocol, so a coding agent such as Claude Code, or any MCP client, can work with the agency's properties, inquiries and viewings directly. It needs a key you choose, and refuses every request until one is set.
Models
Included with your purchase. Sign in to read, or open it in your download.
The providers and models the assistant offers, and how one is chosen.
The tools
Included with your purchase. Sign in to read, or open it in your download.
The 19 tools the assistant and the MCP server share, the permission each needs, and how a turn runs them.
Connecting an MCP client
Included with your purchase. Sign in to read, or open it in your download.
The MCP key, the endpoint, and how to connect a coding agent.