01 Describe the goal
Create a conversation, choose a work folder and an available model, and describe what you want done. Add acceptance criteria to guide the final review. An ordinary folder is enough; no Git repository is required.
02 Choose an execution device
Sending directly queues the task locally. To use another computer, pair the devices first, then select it with @. Each device running Rivloom is a Node and manages its own projects, models, and account quota.
Choose a device. Keep the destination. If the device is busy, your task waits on the device you selected. If it is offline or cannot receive the task, Rivloom explains why instead of silently choosing another device.
03 Follow the original conversation
Check delivery, queue position, and the reason for any wait. Handle AI requests according to your approval mode, and add requirements in the same conversation when the direction changes. Without a fresh receipt, an old status is not shown as live progress.
In the local queue, reorder, hold, resume, or reject tasks that have not acquired an execution slot. Pausing the queue does not stop work already running; use Stop in the conversation for that.
04 Verify, then accept
When AI finishes, the task awaits review. Check the actual files, results, and relevant tests, then select Confirm completion if they meet your requirements. To request more changes, continue the same conversation.
Before you start, read the User guide and understand the execution and security boundaries.