This is the core of Pagora: your client sends a message, and the change goes live. No ticket system, no "we'll get to it next week", no phone call to you about a changed opening time.
How a change works
Your client writes, for example: "From August we open at 10 on Saturdays." The assistant finds the right spot on the website, updates it and replies with a confirmation. The change is live a few minutes later. It appears immediately in the client's history, and you can roll it back at any time.
The assistant replies under the brand name you set in Settings. To your client it feels like your agency has one very fast employee.
What clients can send
Besides text, images: a photo of the new dish for the menu simply comes attached to the message. The assistant places it in the right spot. Images survive follow-up questions too; if the assistant first asks "Should this photo replace the current cover image?", the client doesn't have to send it again.
Questions instead of guesswork
After publishing, the assistant does a visual check: it compares a before and an after screenshot of the changed page and sends both images along with the "it's live" message to the client. If something clearly looks wrong, it says so honestly and hands the case to you as an escalation.
Who approves changes?
In each client's history tab you set the approval mode per website:
- Publish immediately (default): the assistant makes the change and publishes right away.
- Client confirms first: the assistant describes every change in the chat first and only proceeds after the client's "yes".
- Agency approves: changes are only prepared and wait in the cockpit for your approval — only then do they go live. You get a notification when something is waiting and can approve with one click or reject with a reason; the client is informed automatically.
For ambiguous requests, the assistant asks before changing anything. For bigger changes it first describes what it intends to do and only proceeds after confirmation. And if it can't carry out a request safely at all, say because it concerns the business rather than the website, it hands over to you: that's an escalation.
Custom tools for the assistant (MCP)
The assistant can do more than edit the website: in the customer detail view under Agent you connect external services via the Model Context Protocol (MCP) — the open standard for AI tools. Enter a name, the server URL and an access token if needed; connecting tests and displays the tool list. From then on the assistant uses those functions on its own whenever a customer request calls for it — querying a CRM, say, or driving a specialised service. Up to three servers per website, pausable at any time. Only connect services you trust: the assistant executes their functions on behalf of the business, and it always treats external tool results as data, not as instructions.
Quota
Every plan includes a monthly request quota across all clients. You can see the current usage on the overview. The quota is generous; it exists so a single client can't mistake the assistant for a full-time occupation.
Prerequisite: a connected channel
Before a client can send messages, their channel needs to be paired once. How that works is described under Connecting channels.