Operating subagents across long, consequential work will be risky. Parents need to poll the subagent to get the progress, which wastes token, and one interruption like laptop power-off will make the subagent's run state unrecoverable.
So I built a runtime. You can define reusable subagent workflows and orchestrate agent according to the workflow. The runtime will supervise the agent run and store durable workflow states in database, it can handle retry-able errors automatically, you can pause and resume the workflow anytime you want and make your daily workflow easier to operate.
It's open-sourced and self-hosted, also it's launching today, for more information: https://www.producthunt.com/products/oh-my-subagents
ringlochid•1h ago
So I built a runtime. You can define reusable subagent workflows and orchestrate agent according to the workflow. The runtime will supervise the agent run and store durable workflow states in database, it can handle retry-able errors automatically, you can pause and resume the workflow anytime you want and make your daily workflow easier to operate. It's open-sourced and self-hosted, also it's launching today, for more information: https://www.producthunt.com/products/oh-my-subagents