Show HN: Jacquard, a programming language for AI-written, human-reviewed code

Jacquard is a research prototype for running, reviewing, simulating, and trusting programs written by models and reviewed by people.

Concretely, it is a small programming language with a compact .jac

surface syntax, an OCaml checker and CPS interpreter, a C-emitting native AOT backend that currently compiles the kernel .jqd

carrier, a command-line tool, a Jacquard-written standard library, and a test framework called Warp. Version 0.1 works end to end but is a research prototype, not a production language; docs/release/0.1/LIMITS.md

is the honest boundary.

Install the 0.1 release candidate without OCaml or opam:

curl -fsSL https://raw.githubusercontent.com/jbwinters/jacquard-lang/jacquard-core-0.1-rc3/scripts/install.sh | sh
~/.local/bin/jac run ~/.local/share/jacquard/demos/basics/m1-fact.jac

The expected output is 120

. Linux x86-64, macOS Intel, and macOS Apple Silicon binaries are published; development from source is documented below.

Then run one policy under concrete and probabilistic telemetry worlds, followed by sampled and exhaustive Warp checks:

sh ~/.local/share/jacquard/demos/case-studies/release-risk/run.sh

Most languages tell you what a program computes. Jacquard also exposes which effects it may perform, finite discrete uncertainty, and canonical program identity. Tools can inspect all three because they live in the language rather than only in comments, logs, or your memory of the codebase.

Things you can do here that most languages cannot offer:

• Read one line and see the effects a function may perform. A signature like (text) ->{net} text

says the function may perform thenet

effect. The Jacquard runtime rejects unhandled world effects unless their authority is explicitly granted with--allow

, including effects performed by dynamic code. This is language-level enforcement in a research runtime, not a substitute for an operating-system sandbox. - Run one program against many worlds. The same code can run against the real network, a scripted fake, a recording of last week's traffic, or a probability model of how servers usually behave. A handler is the piece that answers a program's requests to the outside world; you swap the handler, and the code never changes. This can replace much conventional mocking at effect boundaries and makes "what would my agent do if the API went down?" an ordinary test.

• Enumerate exact probabilities for finite discrete models. A program can sample weighted choices and record evidence, and enumeration lists every reachable outcome with its exact probability. The repair demo below treats a failing test as evidence and computes which patches remain possible and how likely each is.
• Rename and reformat without changing canonical identity. Jacquard hashes canonical resolved structure rather than source bytes. Comments, formatting, provenance, and ordinary local or term renames are erased; pure tests rerun only when canonical code or dependency content changes. This is structural identity, not a proof that arbitrary programs are behaviorally equivalent.

The bet behind all of this: when most code is written by machines, the humans reviewing it need the language itself to answer "what can this touch, and how sure are we" without reading every line.

Read docs/SKILL.md

first. It compresses the kernel, the CLI, the prelude, Warp testing, and the known gotchas into one file, and it loads as a project skill from docs/SKILL.md

. Operating rules are in AGENTS.md

. What will save you time:

• Behavior is pinned by evidence: cram transcripts under test/cli/

, corpus goldens, demo scripts, anddocs/release/0.1/CLAIMS.md

. If a pin fails, treat it as information about your change, and never weaken a pin to make a diff pass. - The kernel is 27 forms ( docs/ast.md

);.jac

is a projection onto those forms, and bootstrap.jqd

remains permanently supported. Treat the shipped surface boundary and its parked follow-ups as release evidence, not as a frozen grammar; do not add out-of-scope features (AGENTS.md

lists them). - The development gate is dune build @all && dune runtest && dune fmt

followed by a cleangit diff --exit-code

.

For readers who speak programming languages:

• One uniform representation: every form is a (head, meta, args)

triple, and the kernel grammar has 27 forms. Quoted code is ordinary data. - Algebraic effects with deep, multi-shot handlers. A handler can resume a computation zero, one, or many times, which is what makes exhaustive search and exact inference ordinary library code.

• Explicit capability grants. The runtime installs handlers for the outside world only for effects you pass with --allow

; there is no ambient authority. - Type-and-effect rows. Every arrow carries the set of effects the function may perform, so a program's inferred row is its authority manifest.

• Discrete probabilistic programming as a library: sample

andobserve

are effect operations, and each inference algorithm is a handler. - Content-addressed definitions. Identity is a hash of canonical resolved structure with non-identity metadata erased, so formatting, comments, and ordinary local or term renames change nothing downstream.

• Tooling that leans on the above: formatter, structure-aware differ, Warp tests with a content-addressed cache, record/replay, and a reproducible release evidence pack.
• A native AOT path that emits C, specializes and caches units by content hash, and is differential-tested against the interpreter under clang and gcc.

The prototype is complete against its original core plan and has since added the public surface syntax, ringed standard library, Warp properties and cache, native compilation, packaged binaries, and product-scale case studies. The RC1 semantic boundary is pinned by 554 Alcotest/QCheck cases, 32 cram transcripts, 21 documentation examples, native sanitizer/leak/fuzz lanes, and a fresh-clone evidence workflow. RC2 repaired binary-demo packaging; RC3 adds an explicit runtime/output license exception and packages the native runtime. The current successor distribution relicenses Jacquard under Apache License 2.0 and keeps that runtime/output permission as an explicit clarification. These licensing and packaging changes do not change the language semantics pinned at RC1.

Here is one handler resuming one continuation twice. The block is copied byte-for-byte to test/docs-doctest/fixtures/readme-multishot.jac

and run by the documentation test lane:

effect Choice where {
  choose : () -> Bool
}

handle {
  match choose() {
    | True -> 1
    | False -> 2
  }
} {
  | return x -> x
  | choose() resume continue -> add(continue(True), continue(False))
}
bash
$ jac run test/docs-doctest/fixtures/readme-multishot.jac
3

The handler ran the rest of the program once with true

and once with false

, then collected both results. That ability to resume more than once is why exact Bayesian inference is a library handler here rather than a runtime feature. The repair demo builds on it: mutate a buggy program's quoted AST into candidate patches, treat a failing test as an observation, and read off the updated probabilities. Running candidate code is an authority, so the pure step still runs (it counts eight candidate patches) and then the demo refuses until you grant the rest:

$ jac run demos/tooling/repair.jac
8
error[E0814]: this program requires the `eval` effect, which is not granted (performed via `posterior-over-patches`)
  hint: grant it with --allow eval, or handle the effect in the program
$ jac run demos/tooling/repair.jac --allow eval

Under the grant, one failing test leaves two surviving patches: the intended fix at 0.75 and a patch that games the suite at 0.25. Adding one regression test prunes the impostor, and the surviving fix prints as a one-line canonical diff: - sub + add

. See sh demos/tooling/repair.sh

for the full transcript.

Most users do not need OCaml or opam. Install the reviewed 0.1 RC binary with:

curl -fsSL https://raw.githubusercontent.com/jbwinters/jacquard-lang/jacquard-core-0.1-rc3/scripts/install.sh | sh

The installer detects your OS and CPU, downloads the matching archive and SHA-256 checksum, refuses a checksum mismatch, and installs under ~/.local

by default. Make sure ~/.local/bin

is on PATH

, then run:

jacquard --version
jac --version

jac

is the short alias for jacquard

. Both commands set JACQUARD_PRELUDE

from the installed package, so ordinary runs do not need an environment variable:

jac run ~/.local/share/jacquard/demos/basics/m1-fact.jac

Narrative demos ship with launchers that choose the installed binary and prelude automatically. They do not require Dune:

DEMO_ROOT="$HOME/.local/share/jacquard/demos"
sh "$DEMO_ROOT/case-studies/release-risk/run.sh"
sh "$DEMO_ROOT/worlds/agent-dream.sh"
sh "$DEMO_ROOT/worlds/escrow/run.sh"

Use these launchers rather than directly running a probabilistic model or a multi-file entrypoint. The launcher selects infer

where observation requires it and assembles related files in isolated scratch space.

To install under a different user-owned prefix:

curl -fsSL https://raw.githubusercontent.com/jbwinters/jacquard-lang/jacquard-core-0.1-rc3/scripts/install.sh \
  | JACQUARD_INSTALL_PREFIX="$HOME/.jacquard" sh

Set JACQUARD_INSTALL_VERSION

to install a different release tag. Supported binary targets are linux-x86_64

, macos-x86_64

, and macos-arm64

; other platforms currently require the development setup.

Release archives are attached to jacquard-core-*

GitHub releases. Each archive contains bin/jacquard

, bin/jac

, libexec/jacquard/jacquard

, share/jacquard/prelude

, share/jacquard/demos

, the native C runtime, and the license, notice, exception, and trademark documents.

These commands assume a fresh clone and asdf

available for installing opam

. If you already have opam

2.5.x, start at the local switch step. If opam

is already initialized on your machine, skip opam init

.

git clone https://github.com/jbwinters/jacquard-lang.git
cd jacquard-lang

asdf plugin add opam https://github.com/asdf-community/asdf-opam.git
asdf install opam 2.5.1
asdf set opam 2.5.1
asdf reshim opam 2.5.1

opam init -y --no-setup --bare
opam switch create . ocaml-base-compiler.5.1.1 -y
eval "$(opam env)"

opam install --deps-only . --with-test --with-dev-setup --with-doc -y
opam exec -- dune build @all
opam exec -- dune runtest
opam exec -- dune fmt
git diff --exit-code

The switch step compiles OCaml 5.1.1 from source, so expect the first setup to take around ten minutes.

The final git diff --exit-code

is part of the development contract: formatting must leave the worktree clean unless you intentionally commit the formatting diff.

Expected versions after setup:

opam

2.5.1 from.tool-versions

• OCaml 5.1.1 from the repo-local _opam/

switch dune

,ocamlformat

,alcotest

,qcheck

,digestif

,menhir

,cmdliner

,odoc

,utop

, andocaml-lsp-server

fromjacquard.opam

In a new shell inside an existing checkout, run:

eval "$(opam env)"

_opam/

is intentionally ignored. It is a local build artifact, not source.

During development, use the built binary through Dune:

opam exec -- dune exec jac -- --help
opam exec -- dune exec jac -- --version

Many direct CLI commands need the prelude. From the repository root:

export JACQUARD_PRELUDE=$PWD/prelude
opam exec -- dune exec jac -- run demos/basics/m1-fact.jac

The main commands are:

jac run FILE.jac [--allow fs] [--allow net] [--dry-run]
jac check FILE.jac [--print-sigs] [--manifest fs,net,console]
jac hash FILE.jac
jac fmt FILE.jac
jac diff FILE_A.jac FILE_B.jac
jac diff STORE_A STORE_B
jac infer enumerate MODEL.jac
jac infer lw MODEL.jac --seed 42 --samples 100000
jac replay TRACE.jqd PROGRAM.jqd [--fork '1=(response 500 "down")']
jac test TESTS.jac [TESTS.jqd ...] [--exhaustive] [--cache-dir CACHE]
jac build FILE.jqd -o PROG

.jac

is the user-facing surface carrier. Bootstrap .jqd

remains fully supported as the internal/debug syntax, quote notation, and kernel format of record. run

, check

, hash

, fmt

, diff

, infer

, and test

select surface syntax by extension; native build, replay programs, the prelude, and many internal fixtures continue to use .jqd

.

Ordinary programs and demos need only a .jac

source file. Do not hand-author a .jqd

twin unless a conformance test specifically needs to prove that both carriers lower to the same kernel and hash. The paired files retained in the corpus and selected demos are evidence fixtures, not an authoring requirement.

jacquard build

currently accepts the kernel .jqd

carrier and compiles a program and its reachable declarations to a standalone binary whose output is byte-identical to jacquard run

— stdout, stderr, and exit codes, pinned by a differential harness in CI (scripts/native-diff.sh

). The full effect language compiles, including capturing and multi-shot handlers, and code values compile since task 73 — quotes, splices, and the structural code ops. eval

alone stays on the interpreter tier (E1102 policy: dynamically loaded code runs where the authority model lives).

export JACQUARD_PRELUDE=$PWD/prelude
export JACQUARD_RUNTIME=$PWD/runtime
jac build demos/tooling/word-count.jqd -o word-count
echo "some words some" | ./word-count --allow console

Requirements and knobs:

• Release binaries discover their packaged prelude and C runtime automatically. Source checkouts may set the two variables shown above.
• A C toolchain: clang (any recent) or gcc. Tail calls are O(1) stack on every toolchain: musttail on clang and gcc 15+, a trampoline below them (the emitted C is identical either way).
• The binary parses --allow EFFECT

(console, clock, fs, dist, infer so far),--seed N

for the sampling grant, and refuses--infer-cache

and--dry-run

(interpreter tooling) with pointed errors. JACQUARD_STACK_MB

sizes the program stack (default 1024): deep non-tail recursion is real C recursion in this backend.- Compiled units cache under .jacquard-native/

, keyed by content, so an unchanged program relinks without recompiling. - Measured performance lives in docs/benchmarks.md

— nine scenarios with interpreter, native (both toolchains), Python, and hand-C columns — with the claim boundaries indocs/native-compilation.md

(reproduce withscripts/native-bench.sh

).

Start with these from the repo root after dune build @all

. The same scripts also work in an installed bundle without opam or Dune:

opam exec -- sh demos/case-studies/stormglass/run.sh
opam exec -- sh demos/case-studies/release-risk/run.sh
opam exec -- sh demos/basics/m1.sh
opam exec -- sh demos/inference/m3.sh
opam exec -- sh demos/worlds/agent-dream.sh
opam exec -- sh demos/worlds/preflight.sh
opam exec -- sh demos/tooling/repair.sh

What they show:

case-studies/stormglass/

: one checkout policy under simulated network and clock laws, exact incident forecasts, and Warp proofs over all 27 worlds.case-studies/release-risk/

: one release policy under concrete and probabilistic telemetry, plus a Warp safety proof over all 18 worlds.basics/m1.sh

: factorial, multi-shot choice, and gated eval.inference/m3.sh

: one model under exact enumeration and likelihood weighting; same model hash, different inference handler.inference/clarifying-question.sh

: an agent computes whether asking the user a question is worth the interruption (value of information).worlds/agent-dream.sh

: one policy under scripted and probabilistic world handlers.worlds/preflight.sh

: candidate agent plans scored under alternate worlds; the live policy still needs a Net grant after the dreams pass.inference/ambiguity-pipeline.sh

: an extraction pipeline that keeps its uncertainty; the user's click becomes anobserve

.tooling/showcase-warp-tests.sh

: Warp checks for the clarifying-question, dream-mode, and ambiguity demos.tooling/repair.sh

: program repair as Bayesian inference; a bug report is an observation over computed single-edit patches, and the most likely patch prints as a one-line canonical-structure diff.worlds/m4-hostile.sh

: generated-looking code that reaches fornet

; signatures and manifests expose the authority.worlds/escrow/run.sh

: product-shaped generated workflow with manifest, dry-run, Warp tests, fault exploration, replay, canonical diff, and approval by hash.

Demo paths are canonical within the categorized directories; there are no flat compatibility aliases. The full catalog is in demos/README.md

.

All public demo outputs are pinned by cram tests (recorded command-line transcripts that fail on any drift), especially test/cli/demos.t

, test/cli/hostile-demo.t

, test/cli/escrow.t

, test/cli/showcase.t

, and test/cli/repair.t

, test/cli/preflight.t

, plus test/cli/case-studies.t

for the larger applications.

The release-candidate evidence pack lives in docs/release/0.1/

.

To reproduce the release evidence from this checkout:

JACQUARD_RELEASE_REF=HEAD JACQUARD_RELEASE_BASE=738dc8e scripts/release/reproduce-0.1.sh

The script installs dependencies, builds, runs the full test suite, checks formatting, runs public demos, runs gauntlet tests, records jacquard --version

, and writes generated evidence under .scratch/release/0.1/

.

Key release docs:

docs/release/0.1/EVIDENCE.md

: what was built and what passeddocs/release/0.1/CLAIMS.md

: semantic claims mapped to tests and caveatsdocs/release/0.1/REPRO.md

: fresh-clone reproduction stepsdocs/release/0.1/FREEZE.md

: frozen version/hash/store/CLI surfacesdocs/release/0.1/GAUNTLET.md

: adversarial tests present and omitteddocs/release/0.1/LIMITS.md

: explicit non-goals and caveatsdocs/release/0.1/DECISION.md

: release-candidate decision memodocs/release/0.1/RELEASE-NOTES.md

: public RC contents and install command

.github/

: CI, release evidence workflow, and PR template.AGENTS.md

: operating notes for future coding agents.bin/

:jacquard

CLI entry point.corpus/

: conformance corpus and golden outputs.demos/

: runnable examples and product-shaped demos.docs/

: design docs, tutorial, CI/CD, Warp, stdlib, errors, release evidence.prelude/

: Jacquard standard library and effect declarations.scripts/release/

: reproducible release evidence script.spec/

: kernel AST and canonical serialization specs.src/

: OCaml implementation.test/

: Alcotest/QCheck suites plus cram CLI transcripts.jacquard.opam

,dune-project

: package and build metadata.

src/form.ml

,src/meta.ml

,src/span.ml

: uniform triple and metadata.src/reader.ml

,src/printer.ml

: bootstrap.jqd

notation and formatter.src/kernel.ml

: validator and typed kernel AST.src/resolve.ml

: names to content-addressed references.src/canon.ml

,src/hash.ml

: HASH_V0 canonical serialization and hashing.src/store.ml

: object store and mutable name index.src/value.ml

,src/eval.ml

: CPS evaluator and multi-shot handlers.src/types.ml

,src/check.ml

: type/effect inference, rows, manifests, exhaustiveness.src/prelude.ml

: prelude , builtin wiring, and root grants.src/infer_dist.ml

: exact enumeration and likelihood weighting.src/diff.ml

: canonical-structure diff over stores.src/warp.ml

: Warp test discovery, running, cache, and properties.

Read these in order if you are new:

docs/README.md

: documentation index and suggested reading paths.docs/tutorial.md

: runnable user-facing examples.demos/README.md

: demo catalog and what each demo proves.docs/ci-cd.md

: GitHub checks and release evidence process.docs/release/0.1/EVIDENCE.md

: release-candidate evidence overview.

Deeper design references:

docs/whitepaper.tex

: historical initial design thesis, motivation, risks, and related work; its roadmap and implementation-status sections are outdated.docs/ast.md

: kernel AST and metadata/hash contract.spec/jacquard-kernel-ast-m0.md

: kernel source-of-truth spec.spec/serialization.md

: canonical byte format.docs/stdlib.md

: prelude and ringed standard library.docs/warp-testing.md

: Warp testing model.docs/errors.md

: diagnostic catalog.docs/development-plan.md

: original implementation plan.

Before opening a PR:

eval "$(opam env)"
opam exec -- dune build @all
opam exec -- dune runtest
opam exec -- dune fmt
git diff --exit-code

When adding valid corpus files, regenerate golden hashes:

opam exec -- dune exec test/gen_goldens.exe

When touching release-facing demos, claims, CI, or semantics, also run:

JACQUARD_RELEASE_REF=HEAD JACQUARD_RELEASE_BASE=738dc8e scripts/release/reproduce-0.1.sh

GitHub Actions has three principal workflows:

CI / Development gate

: build, full tests, clean formatting, version smoke, and release-doc presence on PRs,main

, andrelease/**

.Release Evidence / Reproduce 0.1 evidence

: release branches,jacquard-core-*

tags, and manual dispatch; runsscripts/release/reproduce-0.1.sh

and uploads transcripts.Release Binaries

:jacquard-core-*

tags and manual dispatch; builds Linux/macOS tarballs withjacquard

,jac

, the prelude, demos, and native runtime sources.

See docs/ci-cd.md

for branch protection recommendations.

Jacquard is licensed under the Apache License, Version 2.0. See LICENSE and NOTICE.

Your programs remain yours. Jacquard claims no copyright in source merely because it is written, checked, interpreted, or compiled with Jacquard. Native executables include Jacquard runtime material, so the runtime and generated-output exception explicitly allows user programs and compiled output to use any license their authors choose, including proprietary licenses. The exception also removes Apache License notice obligations that would otherwise arise solely from embedded Runtime Material in a compiled program. This is a statement of project licensing intent, not legal advice.

The Jacquard name and project identity are governed by TRADEMARKS.md. The code license does not grant trademark rights.

Jacquard core is a research prototype, not a production platform. The .jac

surface is implemented and supported but remains an evolving v0 projection onto the permanent 27-form kernel. Native AOT compilation and C-toolchain optimization ship; a VM/JIT, concurrency, membrane enforcement, continuous distributions, gradients, typed staging, language package management, self-hosting, and formal soundness proofs do not. World grants remain coarse. See docs/release/0.1/LIMITS.md

for the exact Core 0.1 semantic boundary.

opam: command not found

: installopam

with asdf using.tool-versions

, or install a compatibleopam

manually.- Dune cannot find packages: run eval "$(opam env)"

in this shell, then reinstall deps withopam install --deps-only . --with-test --with-dev-setup --with-doc -y

. jacquard

cannot find names from the prelude: setJACQUARD_PRELUDE=$PWD/prelude

or run through Dune from the repo root.- Formatting changed files: run opam exec -- dune fmt

, inspect the diff, and commit the formatting changes if they are intended. - Release reproduction writes generated evidence under .scratch/release/0.1/

by default. SetJACQUARD_RELEASE_OUT

to use another disposable output path.

source & further reading

github.com — original article