Container customisation/approach

[lukemalcolm]

I appreciate the focus on security & sandboxes in moltis. I did have a couple of questions:

• What's the rationale behind setting container images at a session level, rather than an agent level?
• It would be great to get granular and custom with docker images & configs (network, host fs/volume access) per agent. This would allow you to be more restrictive with the capabilities (damage) a single agent can cause. Is that something others have a desire for? Are there particular architectural decisions that have been taken that counter this approach?

[penso]

1. Agents were added after, it could make sense to attach agents to a given container image.
2. Agreed, maybe we should specify all those capabilities per agent, it would be easier.

[lukemalcolm]

While I'm not a Rust dev, I am a C++ one. If you don't think these features are high enough on your todo list, I might have a go at vibing my way to a PR. Let me know.