OpenClaw
AI agents take an online school student from payment to lesson without an administrator
Task
After every payment the online school did everything by hand: check the payment in Zoho CRM, find or create the student profile, grant course access in Kwiga, add the student to the Telegram group, send instructions, process the Zoom lesson recording, upload it to Vimeo and make sure nothing was missed. One student took 7 to 15 minutes, a launch with 20 students took 2.5 to 5 hours. They needed a system that runs this chain on its own and calls the administrator only for exceptions.
What we did
Results
What we did
We built the system on the OpenClaw platform as a set of specialised agents, each owning one part of the process. The CRM Agent checks Zoho CRM every 3 hours for new paid or partially paid deals, validates student data and creates tasks for the next agents. The Access Agent grants access to the right course in Kwiga through its API. The Messenger Agent adds the student to the Telegram group, sends the welcome message and instructions, and notifies the student or the administrator when a manual step is needed; it is written in Python because it drives Telegram through Telethon. The Zoom Agent looks for new lesson recordings daily, the Recording Pipeline Agent downloads them from Zoom, uploads them to Vimeo and sets privacy and allowed embed domains. The Reminder Agent sends lesson reminders, the Weekly Meeting Agent creates the team's Zoom meeting every Monday and posts the link to the chat. Agents exchange structured events and run on a cron schedule on a VPS, with no manual launches and no unnecessary calls to the model.
Approach and results
A separate Audit & Alert Agent checks the whole chain: payment processed, access granted, student added to the group, recording processed, video uploaded to Vimeo, API errors. It reports a failure to the administrator in Telegram before a student notices. The administrator's role changed: instead of routine work they supervise the system and handle exceptions only. Onboarding one student used to take 7 to 15 minutes and now runs automatically; a launch of 20 students takes 15 to 25 minutes of checking alerts instead of 2.5 to 5 hours. Lesson recording processing, 20 to 35 minutes each, is fully automatic. Stack: Node.js for most agents, Python with Telethon for the messenger, Bash for deployment and cron, Zoho CRM, Kwiga, Telegram Bot API, Zoom and Vimeo APIs, Docker Compose and Traefik with Let's Encrypt on a Linux VPS, SQLite and JSON events for agent state.