Terminal: ask an agent to build inside your own CRM
The terminal gives an administrator a conversation with an agent that changes this CRM: objects, fields, automations, reports and your own code. It only ever touches your organization, everything it creates goes away if the organization is deleted, and it is off until someone turns it on.
The terminal is a conversation with an agent that has hands. You describe what you want and it builds inside this CRM: creates objects and fields, writes automations and reports, and can publish your own code. It is the deepest customization tool in the product, and for that reason it is off until an administrator turns it on.
Who can open it
Only administrators, and only when the terminal is enabled for the organization. You will find it in Settings → Terminal. If the item is not there, either you do not administer this organization or the permission has not been granted to your role.
It never touches another organization
This is not a promise, it is how it is built. Your code runs on a separate machine that has no database credentials at all. It reaches this CRM only through the API, with your own organization access, so it sees exactly what your organization sees and nothing else. And if the organization is ever deleted, everything the terminal created goes with it.
Its own window, and it keeps working
The terminal opens in its own browser window so you can keep using the CRM next to it and check what has been implemented as it happens. Closing the window does not stop the work: the session lives on the server, and reopening picks it up where it left off, with the history.
Your code, and the safety net around it
Code published by the terminal answers for a hook: before a record is saved, after it is saved, when a deal is won, on a schedule, or on its own endpoint. Only the hook that runs before saving is allowed to refuse a change, because refusing something already saved would be a cancellation that never happened. An extension that keeps failing turns itself off and tells you why, so your CRM never gets slow because of code inside it.
The AI is ours, and it is measured
The terminal always uses our AI. You cannot point it at your own provider, and that is by design rather than by rule: the machine it runs on never receives an AI key. Everything it thinks is measured against your organization and respects your spend limit, exactly like every other AI feature. There is also a ceiling per session, so an agent that gets stuck in a loop cannot spend a month overnight.
When it needs our code
Some requests cannot be solved inside your organization because they need a change in SellioCRM itself. The agent will not pretend otherwise: it opens a request that reaches our team, and you follow the status like any other request you send us.
What the agent does when you ask
When your request arrives, the agent reads this organization's real structure (objects and fields), writes the code in the isolated workspace, runs it and shows the result. Every step shows up on screen as it happens: the line saying what it is about to do, the tool it used, the file it wrote.
- It never claims to have run something it did not run. With no execution machine configured, it says so, and you see it in writing.
- It reads the real fields before answering about them: field names are never guessed.
- Publishing an extension wires real code into the CRM. The agent only publishes when you ask, and says what it is wiring up; the extension starts switched off.
- The turn has an end: there are step and time limits. On hitting one, it says it stopped at a limit, and you ask it to continue.
- The isolated machine can only reach this CRM's API. Nothing leaves outside that list.