For the builders and founders out there, I am going to share our playbook on how you can scale OpenClaw (it probably applies to Hermes as well) instances for your SaaS, because we are not building in that space anymore

In our first iteration for @QasperAI we were providing OpenClaw instances through our own app. Users could download the mobile app and use OpenClaw as their personal agent, via our own interface. The goal was to democratize Openclaw usage for people who have no idea what hosting and agents are.

There are big 3 problems when you are building in that space:
1. Speed (How quick you provision the environment)
2. Security (How to not lose everything)
3. Memory (Store user data across sessions)

I will break down each of them below 1. Speed

This was one of the bigger bottlenecks. An OpenClaw instance needs around 6-15 seconds to start from a cold state (it depends on where you are hosting your app)

That is a very long time for a user to stare at a screen. In order to provide a good UX, this needs to be done in less than 3 seconds, since you can probably hide 1-2 seconds via other tricks.

The way we solved it was with warm containers. We used Docker for our implementation, but there are better approaches out there.

When the app starts, we created 8 warm containers, they contained no credentials, new instances, nothing. Just an OpenClaw instance. When a user logged in, they would claim that container and we would assign it to them. In the background we would create another warm container, trying to keep our warm container count always at 8