Envelop

Deterministic by design.

Envelop runs without AI. It was designed as deterministic software around Tomato’s open-compute philosophy: reviewed rules select source-linked answers, while the compiler exposes the program, bytecode, target, and result provenance.

Same inputWhat is (57 + 19) AND 0x3F?

Generative response space

many possible outputs

A generative stack maps the input through learned weights or a remote inference service and selects from a broad response space.

Envelop’s route for the same task

  1. Input text
  2. Reviewed rule
  3. Compiled Tomato program
  4. Labeled result12 · 0x0000000C

One inspectable pathThe behavior catalog, router, and help stay within a tested 96 KiB source budget, with 0 bytes of model weights.

The same input, state, and version take the same route. Open compute makes the route and evidence inspectable, from a matched answer rule to the bytes Tomato executes.