Issue
It would be really cool if we could use Aider for all sorts of system ops work as well as for writing code. To let an LLM lose in my system, I would require quite tight constraints on what I give it access to. One way to achieve that is to run aider in a docker container where only specific host directories or files are mapped into the container. That way you know exactly what the LLM has access to and gain some measure of control.
An issue with this is that you can only set what directories are mounted to the container when launching the container, you can't change bind-mounts on a running container. This means that the wrapper needs to be able to restart aider with new directories mounted from time to time, letting the user change what directories aider has access to. To restart aider transparently it needs to be possible to snapshot all it's state and reload it on startup. That probably requires quite a bit of work.
Version and model info
No response
Issue
It would be really cool if we could use Aider for all sorts of system ops work as well as for writing code. To let an LLM lose in my system, I would require quite tight constraints on what I give it access to. One way to achieve that is to run aider in a docker container where only specific host directories or files are mapped into the container. That way you know exactly what the LLM has access to and gain some measure of control.
An issue with this is that you can only set what directories are mounted to the container when launching the container, you can't change bind-mounts on a running container. This means that the wrapper needs to be able to restart aider with new directories mounted from time to time, letting the user change what directories aider has access to. To restart aider transparently it needs to be possible to snapshot all it's state and reload it on startup. That probably requires quite a bit of work.
Version and model info
No response