Yes — the ultimate goal is to develop an interpreter that includes a complete game engine, so it might look that way. I have seven years of programming experience, so I can both write and read code; I verified the LLM-generated code and fixed bugs by comparing it with code on GitHub. I actually wrote most of the PHP side myself. I appreciate the suggestion and will try the method you mentioned, but I found it faster to read the code myself to find and fix errors directly (the shader parts couldn’t be solved by the LLM, so I’m studying tutorials and implementing them by hand). I think that for people who didn’t major in programming, it’s still probably difficult to have an LLM produce even basic working programs.
Wow, this is super impressive — the breadth of tools you’ve packed into Context Sync is amazing.
I really like how you’ve organized everything from core memory to advanced analysis and even cross-platform sync. The local-only storage design is also very reassuring.
The AI runtime part is still pretty experimental — right now it’s basically a set of hooks that let me call out to external processes (via FFI) and experiment with model inference or code-gen helpers. The idea was to see how far I could push PHP into interacting with AI runtimes without leaving the framework context.
Context Sync sounds really interesting. Persisting framework decisions across AI conversations could make the dev workflow much smoother. I’ll definitely take a look at how it might connect with Onetone.
Thanks again for the thoughtful comment, and have a great day!