Two small annoyances cost software teams a surprising amount of time. Signed or hashed JSON breaks because two services write the same data in a different order, so the bytes differ and the hash doesn’t match. And CI pipelines redo work whose inputs haven’t changed at all, minute after minute, run after run.
BXM is two small open-source command-line tools built on one idea: same input, same bytes, same result. BXM Canon gives every value exactly one binary encoding, so hashes and signatures don’t depend on who wrote the bytes. BXM Memo skips a command when its inputs haven’t changed and tells you exactly why when they have.
The Pitch in Four Numbers
- 2,000,000 random and damaged byte strings fed to the Canon decoder, and none accepted in a second form.
- 60 conformance vectors in the Canon spec, for anyone writing another implementation.
- 61 milliseconds for Memo to restore 200 files from cache, for a step that took 2.6 seconds to run.
- 16 processes sharing one Memo cache at once, each getting the right result.
BXM Canon
Canon is a binary encoding in which each value has exactly one valid form. The whole format fits on two pages. Two JSON files with the same data in a different key order encode to the same bytes, so they hash to the same value and a signature on one is valid for the other.
It isn’t a replacement for JSON. JSON stays the readable form, and Canon reads and writes it. Canon is for the moments when exact bytes matter: hashing, signing, deduplicating and comparing. The case studies show why that matters. A 128-line config diff hid two real changes, and Canon named them in two lines. Standard deduplication of 100,000 JSON lines found no duplicates. Hashing them through Canon found 9,900.
Canon has one Rust core, compiled to WebAssembly for a JavaScript package that runs in Node and browsers, and wrapped for Python with wheels for Linux, macOS and Windows.
BXM Memo
Memo wraps any command. You declare what a step reads and writes. The first time, it runs the command and saves the outputs. The next time, if the inputs are the same, it restores the outputs instead of running anything. If it can’t skip, it tells you which input changed. In the site’s example, one run misses because an input changed, and the next hits and restores 200 files.
Memo doesn’t need a build system. Existing scripts, Makefiles and CI steps stay as they are. It’s a single Go binary using only the standard library, it has a GitHub Action, and parallel runs writing the same output folder are handled safely.
Who It’s For
- Teams signing or hashing structured data, from API payloads to audit records and software manifests.
- Anyone deduplicating or comparing JSON who keeps getting fooled by key order and formatting.
- CI owners paying for minutes spent redoing unchanged work.
- Developers with slow scripts that don’t live inside a fancy build system.
Limits
Canon isn’t for storage or display. Memo isn’t a build system and only knows about the inputs and outputs you declare, so a step that reads something you didn’t list can be skipped when it shouldn’t be. The two are separate binaries, and you only need the one that fits your problem.
The Four-Question Screening
Every project featured here gets the same four questions, the way you’d screen a startup idea.
1. Where is the money in it?
Both tools are free under Apache 2.0, with no paid tier. Memo has the clearer money: CI minutes are a real bill, and caching that cuts them pays for itself. A hosted, shared cache for teams, with reports on time and money saved, is the obvious paid product, and a well-trodden one. Canon’s value is in trust infrastructure, signed records, supply-chain metadata, audit logs, where companies pay for libraries with support and long-term stability guarantees across languages.
2. What real problem does it solve?
Two kinds of waste. Hashes and signatures that fail on data that’s actually the same, which causes bugs that are maddening to track down. And compute spent redoing work nobody changed, which costs time on every push.
3. The technology in plain words
Canon is like insisting that every address is written in exactly one format: same order, same abbreviations, same spacing. Then two letters to the same place look identical and you can tell at a glance. Memo is like a kitchen that remembers every dish it’s cooked. If you order the same thing with the same ingredients, it hands you the one already made, and if something’s different, it tells you which ingredient changed.
4. Moat and differentiation
Canonical encodings exist, and so do build caches. Canon’s edge is a tiny spec with a strong test record, two million hostile inputs without a second form, and one core shared across Rust, JavaScript and Python. Memo’s edge is that it works with whatever you already have, with no build system to adopt, and explains every miss. The moat that can grow is adoption: an encoding is worth more the more places use it, and a cache becomes sticky once it’s in the pipeline.
Why It Matters
Repeatability underpins trust in software: the same input should give the same bytes and the same result, every time, on every machine. BXM makes that easy in two places where it usually isn’t.
Follow the Project
Read about Canon and Memo, browse the use cases, start with the quick start, or get the code from the BXM repository on GitHub. BXM is listed with the other open-source projects on SoftDevStudio.com, and more featured projects are in the Projects section.