slopstack-labs/sloppiler

Sloppiler is a joke project, written in Go, that replaces a traditional code compiler with an AI language model. A normal compiler takes source code and converts it to a runnable program through a series of well-defined steps: parsing, type-checking, optimization, and so on. Sloppiler skips all of that and asks an AI model to produce binary output directly, describing the intent of the source code as best it can. The predictable result is that the output almost always crashes immediately when you try to run it. The README tracks this openly in a "track record" table showing that nearly every combination of model and mode produces a segmentation fault, which is what happens when a program tries to access memory it is not allowed to touch. One combination did work: a local model called codellama in "optimistic" mode successfully printed "Hello, world!". Most others did not. The project supports several AI providers: a local model via Ollama, OpenAI, Google Gemini, and Anthropic Claude. You point it at a source code file, choose a provider and model, and it attempts to produce a runnable binary. The README includes provider and model recommendations written in the style of serious engineering documentation, though the recommendations are part of the joke. Two additional modes exist beyond the basic one. An "optimistic" mode routes the AI output through an assembly tool chain. A "loop" mode feeds compiler errors back to the model for self-correction cycles. A "force iterate" mode runs multiple improvement passes even when a binary is produced without errors. The README is written as parody corporate tech documentation, using buzzwords like "inference layer", "stakeholder experience", and "binary ideation". The project is openly self-mocking, calling its own output pipeline "blazing-fast time-to-segfault" and noting that the only compiler insight it is built on is that code does not need to be understood, it needs to be shipped.