Connect ChatGPT to your business software, with a confirmation before every change

A connected assistant gets 81 actions across 9 Orbiz tools. 36 of the actions in the catalogue change something; the rest only read. None deletes anything, and the Vault has no actions at all. The work people hand over most often looks like this:

• Turn meeting notes into Cadence issues, move them across the board, and close them with a comment.
• Log hours against a project and milestone in Trackline, and total what you logged this month.
• Check your leave balance in AMS, request a day off, or approve the requests waiting on you.
• List open jobs and candidates in HRMS, add a candidate and move one to the next stage.
• Answer “what was my net pay last month?” from Ledgr, for the person asking.
• Find a file in Filespace, save a summary as a text file and share it.

What an assistant can and cannot reach differs by tool, and each tool’s page spells it out. For the areas people ask about most, see the pages on the HR MCP server, the payroll MCP server, the time tracking MCP server and the project management MCP server.

ChatGPT connects to a service like Orbiz through what OpenAI calls developer mode. OpenAI’s guide describes it as full MCP client support for all tools, both read and write, and lists it for Pro, Plus, Business, Enterprise and Education accounts on the web. You switch it on in ChatGPT’s settings, then create an app that points at a remote MCP server: in this case, your organisation’s Orbiz address.

OpenAI is blunt about the trade-off, and it is worth repeating here. Its guide calls developer mode powerful but dangerous, and names the risks: prompt injection, a model’s mistakes on actions that write, and servers that are not what they claim to be. None of that is specific to Orbiz, and all of it is a reason to connect only services you trust and to read what an action is about to do.

The first check is ChatGPT’s. By default it asks you to confirm an action that writes, and shows what it is about to send. You can let it remember your approval for the rest of a conversation; a new conversation asks again. Read the confirmation before you accept it, especially for anything created in bulk.

The second check is Orbiz’s, and it happens before ChatGPT is involved at all. On the approval page you tick what ChatGPT may read and change, tool by tool, and only what your role could already do is offered. If you tick reading alone, ChatGPT has no way to change anything in Orbiz, whatever it is asked. Changing pay is a separate tick that carries its own warning.

ChatGPT’s deep research works with a connected service through two read-only tools, search and fetch. Orbiz provides both. That lets a research task draw on what your organisation keeps in Orbiz alongside the web, for example “pull together what we have on the Northwind project”.

Search and fetch read; they change nothing. And they run as you, like every other action, so research cannot surface something your role cannot open in Orbiz.

Every action ChatGPT takes in Orbiz is listed in an activity log with the outcome — done, not allowed or failed — and the time. Each member sees their own in Settings → Connected AI apps; admins see the organisation’s, with the member’s name on each row. An admin can disconnect any connection and can switch AI assistants off for the whole organisation at once. The platform’s wider safeguards are on the security page.