one feature we were missing though, was the agent being able to log into services. anthropic understandably blocks this capability with their guardrails, and you wouldn't want your credentials to end up in any model context anyways.

so we added a keychain service to the vm that the agent is using. it was built using the pass cli (https://www.passwordstore.org/). it's not perfect yet but already works solidly (feel free to contribute any improvements).

take a look at the implementation here (https://github.com/cuse-dev/cuse/blob/main/platforms/linux/a...), all feedback is highly appreciated. if you want to get going asap, here is a quick nextjs quickstart repo (https://github.com/cuse-dev/cuse-quickstart), if you want to take a look at the whole sdk, see the documentation (https://docs.cuse.dev).