What is muforth?

muforth is a small, simple, fast, indirect-threaded code (ITC) Forth intended for use as a cross-compiler for microcontrollers and other embedded devices. It is written in C and its core is very portable. Because of its Forth nature, it is naturally extensible, scriptable, and customizable.

muforth is very well-suited to interactive coding, debugging, and exploration, and is a great tool for bringing up new hardware.

It has support – in varying degrees of completeness – for a number of different architectures and chip families; see the Github README for the current status of all the targets.

Tethered vs self-hosted

Unlike a “self-hosted” Forth, where the target contains the dictionary, the text interpreter, and all the tools necessary for assembling and compiling code, muforth supports a tethered development model. muforth runs on the host machine, compiling code and data that are later copied to the target machine and executed there.

Because the dictionary and all the compilation tools reside on the host, only the compiled code and data reside on the target. This makes it possible to target very small devices that only have a few kilobytes of flash and a few hundred bytes of RAM. In contrast, a self-hosted Forth often needs 16 KiB (or more) of flash to be useful, and consumes RAM for the text interpreter and compiler.

The host machine is also orders of magnitude faster than the target, so doing the compilation on the host is essentially instantaneous.

Why yet another Forth?

I initially wrote muforth because I wanted to try out some implementation ideas. The core language primitives are written in C, but initially muforth compiled Forth words via a simple native code compiler for the x86. I quickly realized that simplicity and portability were more important than speed. The current implementation is a straightforward indirect-threaded Forth – and it is quite fast!

Its implementation is no longer the point. Its sole reason for existing is to be a cross-compiler for other Forths, and their implementations are in no way tied to muforth’s. In fact, muforth can be used to compile any sort of code onto the target. I’ve used it very successfully as a smart assembler for writing AVR and S08 code.

By keeping it small and simple, it is much more likely to be a useful tool that people can customize.

It’s BSD licensed, so do what you want with it! I’d love to hear about innovative or unusual uses of muforth.

Starting points

The code is hosted at Github.

BUILDING will tell you how to build muforth. It’s very simple.

Look in mu/target/ to find a target that interests you. There is generally a mu/target/<target>/build.mu4 that loads the cross-build environment. Use it as an “index” to find the assembler, disassembler, meta-compiler, kernel, and other key pieces of code.


Sadly, there isn’t a lot of documentation right now. A good place to start is to read the source.

The code – both the C code that implements the Forth virtual machine and the muforth code that does everything else – is carefully and in some cases extensively (even obsessively!) commented. Right now your best source of documentation is the code! Luckily for you, there isn’t that much, so reading it is actually possible. That’s part of the point of muforth; I want it to be a convivial tool.

The key to the system is the Forth code that muforth reads when it first starts: mu/startup.mu4. You’ll learn a lot by reading this!

The index of all pages is also a place to look. I’m planning to write a manual for the overall system and “getting started” guides for each of the targets.

NOTE: The state of muforth’s documentation will get much better! Especially if I hear feedback about how terrible it is. ;-)

News and history

The muforth blog is made up of yearly journal pages. Read them here:

The original README – though outdated in many ways – is a good introduction to why I wrote muforth and what I was trying to achieve. It describes the native-code compiler version of the system. That, and a few other things – notably create/does, the tokeniser, and the lack of tail-recursion in the threaded version – have changed since the README was written, but much of it is still true, applicable, and hopefully of interest.


In March 2008 I gave a talk about bootstrapping, muforth, and convivial tools.

Warning: I wave my arms around a lot, and the audio and video quality isn’t that great, but you might find it interesting, or at least amusing.

It’s also hard to see my slides. If you want to “follow along”, download my slides, and use less to view them – ideally in a text window that is at least 30 lines high – like so:

  less -30 ~/muforth/talks/2008-mar-30-PNCA

Above all, enjoy!