neural/rune/docs/explanation/instant-start.md

instant start

Between a rune program becoming available and it starting to run, only one thing happens: parsing. There is no compilation phase.

what it means

source ──parse──> AST ──lower──> Nox noun ──reduce──> result

Each arrow is milliseconds for any human-scale program. The interpreter walks the Nox noun and reduces it directly; it does not first emit bytecode, build a type-checked IR, or run a static analysis pass. The moment the source is in hand, it runs.

why it is non-negotiable

rune is the authoring layer for the cybergraph — agent kernels, terminal programs, dialects evaluated on demand. These need to react to events with low latency and reload code by rewriting their own subject. A build phase between source and execution would poison all of it: every hot reload, every dynamically-built formula, every REPL keystroke would pay a compile tax.

So every architectural decision is checked against one question: does this preserve the ability to run a program the moment its source becomes available, with no phase between? If a feature requires a build step, it does not belong on the default path.

the compile tier does not break it

rune has a compile back-end (rune-compile, milestone M6) that lowers hot functions further for steady-state speed. It runs in the background, never on the path to first execution. A function is interpreted instantly on its first call; only if it runs hot does compilation happen, off to the side, and the compiled form is swapped in on a later call. The interpreter path stays the contract; compilation is an optimization layered on top, never a precondition.

see also

Graph