Skip to content
agncrowPublic

About

A programming language for data work and web applications: looks like Python, compiles to Rust, runs parallel by default, treats errors as values. One binary, two execution modes held to one behaviour.

Topics

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Repository files navigation

Crow

Crow

A programming language for data work and web applications. Looks like Python, compiles to Rust, runs parallel by default, treats errors as values.

Alpha. Crow is open source under the MIT or Apache-2.0 licence, at your option (LICENSE-MIT, LICENSE-APACHE). The language still changes between releases (CHANGELOG); it takes no pull requests during the alpha, and bug reports are welcome in Issues (CONTRIBUTING).

use math

prices = [101.5, 102.1, 99.8, 103.4]
avg = mean[prices]

returns = for p in prices
    (p - avg) / avg

log['average: `fmt[avg, decimals = 2]`']

An operation that fails answers a value that says why, and the program decides what to do with it:

prices = read['prices.csv']
if not prices.ok
    log['no prices today: `prices.reason`']
  • Parallel by default — for loops run their iterations concurrently; no threads, no async
  • Errors are values — no exceptions, no null; failures carry .ok / .reason and refuse to be silently misread
  • One way to do things — one call syntax f[x], one string quote, one loop; every rejected feature has a documented reason
  • Batteries included — tensors, data frames, SQL, a reactive web framework (LiveView-style view blocks), in one binary; language models — a model as a connection, and an agent loop over the program's own functions with a gate, skills and MCP — are Raven's packages, written in Crow
  • Verified — crow check runs your tests plus generated edge cases, replays recorded database and model answers, and diffs what the app does against the accepted baseline; the two execution modes are held to one behaviour on every block of the book
  • Held to references — frames to pandas and models to scikit-learn to the last digit, optimisation to scipy step for step, retrieval to rank_bm25 and faiss, a GPT to nanoGPT; each package ships with the reference script it is held to

Quick start

One line on a Mac or a Linux machine (on Windows, use WSL):

curl -fsSL https://raw.githubusercontent.com/agncrow/crow/main/install.sh | bash

It puts one prebuilt binary in ~/.crow/bin, checks it against its published hash, runs a hello program, and prints the PATH line. Nothing else is installed: the interpreter, the tests, crow check, the language server and served apps work at once.

crow new hello && cd hello && crow run

Compiling to a native binary needs a Rust toolchain. The first crow build fetches a pinned one into ~/.crow, with the runtime, once (about 175 MB); crow setup does it up front, and crow setup --check says what is present. Until then crow run says so and interprets, which answers the same, only slower.

crow help lists the main commands, and the toolchain guide every one. To upgrade, crow upgrade replaces the binary with the latest release, held to its published hash; to remove Crow, delete ~/.crow and the PATH line. Working on Crow itself? git clone https://github.com/agncrow/crow && cd crow && ./install.sh --dev builds your checkout in place; --source builds from it and snapshots the source for compiled programs.

Learn

  • The Crow Book (source) — step-by-step, from install to a live dashboard. Every code block is executed in both execution modes by CI (crow test --book).
  • Toolchain guide — every crow command, project layout, the crowfile.
  • Run the book's examples in place — from a checkout, crow book serves the book with ▶ Run / ⚙ Compile / ✓ Test / ■ Stop on every code block.
  • Learning with an AI assistant? Paste docs/crow-for-llms.md into it and ask it to teach you interactively — the file is written to make any LLM an accurate Crow tutor.
  • Language specification — the reference, including the Design Notes explaining every deliberately rejected feature.
  • Runnable programs in examples/ (indexed in its README), and the full documentation map in docs/README.md.

Two execution modes, one behavior

crow run -i file.co interprets instantly; crow run file.co transpiles to Rust and runs a native binary. The two modes are held to identical output on a growing parity corpus (crow test --parity). Interpret while exploring, compile when shipping.

Licence

Crow is licensed under either of the MIT licence (LICENSE-MIT) or the Apache licence, version 2.0 (LICENSE-APACHE), at your option. The runtime crates a compiled program links are under the same terms, so a program built with Crow carries only its author's licence.

About

A programming language for data work and web applications: looks like Python, compiles to Rust, runs parallel by default, treats errors as values. One binary, two execution modes held to one behaviour.

Topics

Resources

Contributing

Stars

0 stars

Watchers

0 watching

Forks

Releases

Packages

Used by

Contributors

Languages