Sandbook env for code execution?? Free options

building an rl env that needs a sandbox env for running code. the possible choices are PrimeIntellect modal e2b but first thing that would take credits & get exhaust pretty quickly ig, also alibaba opensandbox if we deploy that in hugging face spaces that would cause a docker in docker issue.
so using subprocess is a risky but worth considering. even the test code ran easily. so is there any else approach that i can use??