You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
I think custom code evaluators are one of the nicest feature we have in agenta. It allows much more flexibility and allows to check for specific business rules, computing bleeding edge metrics, etc... However, developing them with the limitations of RestrictedPython turns an exciting opportunity into a chore and severly restrict what could be achieved.
It would be really nice to have the option to opt out from RestrictedPython during the init process (or a developper mode in the UI). In my case, being a solo dev working on agenta on a machine i set up myself, there is no real benefit from trying to sandbox (which I'm not sure restrictedpython even does.). We're simply writing tests against some string inputs in a docker container.
I understand however the celery-worker running the evaluations mounts the docker.sock, which is a big security risk if we begin arbitrary code on it. An even better solution, could be to properly sandbox the code evaluation in its own dedicated docker, but that's more work and potentially more resource usage and latency.
Let me know what you think !
The text was updated successfully, but these errors were encountered:
I think custom code evaluators are one of the nicest feature we have in agenta. It allows much more flexibility and allows to check for specific business rules, computing bleeding edge metrics, etc... However, developing them with the limitations of RestrictedPython turns an exciting opportunity into a chore and severly restrict what could be achieved.
It would be really nice to have the option to opt out from RestrictedPython during the init process (or a developper mode in the UI). In my case, being a solo dev working on agenta on a machine i set up myself, there is no real benefit from trying to sandbox (which I'm not sure restrictedpython even does.). We're simply writing tests against some string inputs in a docker container.
I understand however the
celery-worker
running the evaluations mounts the docker.sock, which is a big security risk if we begin arbitrary code on it. An even better solution, could be to properly sandbox the code evaluation in its own dedicated docker, but that's more work and potentially more resource usage and latency.Let me know what you think !
The text was updated successfully, but these errors were encountered: