In this article i've proposed a simple way to work on multiple tasks in parallel using 1 OpenClaw agent.

The trick is to create multiple groups in Telegram, add you agent there with require_mention=false and work in them.

This way, context for each session is stored separately and you don't need to way for a result on one task to work on another one, while memory is still unified, which allows to cross reference conversations.

Curious how you're handling context bleed between tasks — that's usually where parallel agents fall apart fast.