How it works
Ask in plain language
Tell the assistant the outcome you want — enrich these leads, compare last week’s ad creatives, draft the follow-up sequence. It plans a step, calls a LeadRail capability, reads the result, and continues until the task is done.
See every step it takes
Each step appears as it happens, in plain language, with a live indicator that resolves to a check or a cross. There is no hidden work: if the assistant read your campaign list, you saw it read your campaign list.
Approve before anything leaves
Anything that spends budget, reaches a third party, deletes irreversibly, or switches on a standing rule stops and waits. You see the exact action and arguments, then approve or cancel. The gate is enforced on the server.
The approval gate, in detail
Every capability in LeadRail declares what class of action it is. That declaration — not a prompt, not a setting a model can talk its way past — decides whether it runs on its own.
- Readruns immediately
- No mutation. Runs immediately.
- Internal changeruns immediately
- Changes only LeadRail’s own records. Runs immediately.
- Spendneeds approval
- Consumes credits or ad budget. Approval required.
- External sendneeds approval
- Reaches a real third party. Approval required.
- Destructiveneeds approval
- Irreversible deletion. Approval required.
- Standing ruleneeds approval
- Switches on something that will act repeatedly on its own. Approval required, and the request has to state the ongoing nature and the cap.
- The gate runs on the server, so a modified browser client cannot skip it.
- The approval is written down before the action runs, and the arguments that were approved are the arguments that execute.
- If the proposal changes after it was raised, the approval is invalidated and has to be granted again.
- External tools reaching LeadRail through its machine endpoint get the same gate, and leave the same audit record.