Spent some time on Saturday working on the first exploration into a segregated environment for Firmware compilation and execution using a Docker container. I wanted to get a feel for how difficult it might be to have a LLM work with the tooling. So to start I decided to give an open source tool a shot, in this case that tool was Open Interpreter. I was pleasantly surprised to find that it worked pretty well after some initial debugging of my setup.
With a single prompt you can have the LLM add a debug message to the EDKII firmware, compile it, and execute the necessary QEMU command to attempt to boot the firmware. The reliability is extremely low, but this is a single shot prompt (with the help of the open prompt auto-run and reprompt behavior that is built in). Another note is that QEMU had no debug output so even though the LLM had successfully invoked QEMU it would never return as it just booted without debug and waited… Not ideal.
I am pleasantly surprised by the promise of a system that took me an hour or so to setup, though like everything it would need some tuning. For the final approach though I don’t think reprompting an LLM for compile and run commands is the right way to go, that should be in the setup phase of any new hardware and left alone. The options open to the LLM should simply be ‘build’ and ‘run’ with a response from both (pass/fail and logs).
Next I think I will peek into Renode and see what that looks like.
Thanks 🙂
-Travis