Jacquard

Jacquard addresses the problem of trusting code you did not write: it gives reviewers a language-level view of what a program does, not just what it says. The core workflow is writing or receiving .jac source files, running them through the OCaml checker to surface effects and authority grants, and then executing or simulating them through the CPS interpreter or the C-emitting AOT backend. The Warp tooling runs programs against multiple simulated or real worlds, turning ‘what would this do if the environment looked different’ into a concrete, repeatable test.

The differentiating feature is effect tracking at the language level. Rather than auditing output after the fact, Jacquard’s type system encodes what effects a program is permitted to produce — so a reviewer can inspect authority grants and decide whether to approve them before execution, not after. This is not a linting pass on top of another language; it is baked into the type system and interpreter.

Jacquard fits teams doing research into ML-written code pipelines, academics studying capability and authority models, or engineers prototyping a review layer they plan to build infrastructure around. It does not fit teams that need a stable, supported production runtime. The contributor base is small (the repository shows no open issues or pull requests at time of scrape, and 73 stars), documentation is described as a ‘human-friendly introduction’ rather than a reference manual, and there is no indication of a paid tier or commercial support path.

The project is implemented in OCaml and uses a dune build system. Binaries and an install script are provided. The repository includes a bench directory, a spec, a corpus, and demos — suggesting it is far enough along for experimentation but positioned explicitly as a research artifact, not a production dependency.