baskduf/FableCodex

FableCodex wires together two things that most coding agents treat as separate: a structured planning pass borrowed from Fable and a Codex-style execution loop that turns that plan into working code. The idea is that the agent reasons about what it is doing before it starts writing, rather than generating and patching in a tight loop until something compiles. That framing is not new, but most implementations leak at the seam between planning and execution — the plan becomes decorative the moment the model hits a tricky function. The repo is early but the architecture is legible, which is the real test for something you might want to extend or embed in your own toolchain. No hosted demo, no pricing page — this is a raw open-source project that rewards a founder who wants to understand the wiring before the abstraction hardens. Reservation: documentation is sparse and it will take real effort to run against a non-trivial codebase. -> Best for: AI engineer building or evaluating coding agent pipelines