Rolled oats. Flour. Sugar. Golden syrup. Butter. Combine these and you have a pretty delicious batter for a biscuit. In this case, it’s essentially an Anzac biscuit but the same ingredients will get you close to a flapjack, some form of muesli bar or toasted granola for breakfast. Perhaps this niche ingredient list narrows the options considerably, but consider butter, garlic, onion, vegetables, chicken, eggs, milk and flour. The options for recipes from that list are considerable. So what is in a recipe? Fairly obvious right—it must have the instructions and ingredients needed to replicate the picture on the front. Instructions without ingredients aren’t too helpful, and ingredients without instructions are just a shopping list.

And what does this mean for open source software? Well, let’s explore some history of the topic and why it’s important for agriculture—after all farmers as co-developers is a goal for many in this field.

Welcome

It seems simple in food, yet in the open-source community there is an over 40 year history on what constitutes free and open-source, tangentially the definition of a recipe. While definitions exist (see the Open Source Initiative The Definition 1.0) and are an important yardstick, it’s easy to miss the forest for the trees, crop for the weeds or project for the commits, if we stick with strict definitions.

And with that, welcome to the third edition of the OpenSourceAg newsletter. In here I want to explore the history and definitions of open-source software—I mentioned strategy last edition, but I think it’s important to understand the context. It sounds dry, but I promise it is a genuinely fascinating story with some interesting short essays to read, particularly looking back on the more than 40 years of transformational development that has occurred since.

There’s also the latest YouTube video on how to install the OpenWeedLocator software and a quick overview of some work in the pipeline for the project. So, let’s get stuck in.

Table of Contents

• Welcome

• An evolutionary arms race

• Interesting Reads

• OpenWeedLocator Update

  • In the pipeline

• Final Thoughts

An evolutionary arms race

If you listen to some of the excellent podcasts on agtech such as the Future of Agriculture or The Modern Acre, you often hear founders and CEOs mention frequent farm visits and farmer feedback as critical elements of their development cycle. Clearly, understanding your user needs is essential and being on farm is the most effective way learn and test hardware/software in real world environments. But what if this was an inefficient and a slow approach for innovation? After all, these visits require staff and hardware to be co-located with the region of interest, narrowing your development and testing opportunity. Software-only solutions are slightly less bound by this problem.

This challenge is central to Raymond’s quote above—farmers as co-developers is the least hassle route to advancing technology. Eric Raymond’s contributions (such as Cathedral and the Bazaar) in the 1990s and 2000s were formative in defining what was meant by open source, and convincing businesses of its benefits. In the same essay, Raymond expands on the importance of the approach, labelling this discrepancy in efficiency as part of an evolutionary arms race stacked against closed development.

We are seeing this play out now with the LLM space—OpenAI as a closed-source approach is competing heavily on many fronts with open source developers building off DeepSeek, Mistral and LLama among others. The new Le chat from Mistral is even threatening the position of ChatGPT on various App store leaderboards.

DeepSeek on HuggingFace has been downloaded almost 3M times. Think about the opportunity presented by running such a model at the edge. While OpenAI APIs would require a fast internet connection, actually hosting the model on a small computer fitted to tractors would ensure LLM performance even when the signal drops out. This has already been done on a Raspberry Pi by Jeff Geerling (albeit only 1.2 tokens/second).

The point of pursuing open source development in agtech is about advancing this rate of organic improvement, by enabling technology to meet the incredible array of niche use cases that make the industry so interesting. Perhaps this is somewhat controversial, but in agtech we shouldn’t take a binary approach dictated by definitions, but instead be welcoming of approaches to open innovation that meet the ‘vibe check’ of open source. Of course, the OSI definition is the gold standard and somewhere to aim, but even researchers in the field rarely meet this definition.

For example, the Right to Repair equipment is important for farmers (and all consumers) to ensure smooth operations under all conditions, particularly during time-critical periods with peak stress on equipment (generally harvest and seeding). Yet, Right to Repair while passing my vibe check for open source, wouldn’t fall under a strict OSI definition. Richard Stallman, a father of open source, though someone vehemently opposed to calling it as such, advocated strongly for the sense of ‘freedom’ around running, studying, and altering software.

This ‘freedom’ is what leads to innovation, because it enables the pursuance of undiscovered opportunities that do not appear in companies and require an in depth knowledge of the system in which it will be used.

I will leave it at this for this edition—it has been quite high level and less applied open source, but the more I read the more interested I became, so though I should share it with you all.

Interesting Reads

This week the interesting reads are some early essays from people involved in the ‘free software’ and ‘open source’ communities. Some have been referenced above.

Cathedral and the Bazaar is a particularly interesting read. Eric contrasts the old-school, top‑down approach to software development—cathedral—with a dynamic, community‑driven process of a bazaar.

1. Cathedral and the Bazaar | Eric S. Raymond

2. Yes, give it away. | Richard Stallman

3. Producing open source software | Karl Fogel

This one is a 190 page behemoth, but it is a fantastic resource for those interested in definitions, strategies and tools for building an open-source project and community. Possibly one of my favourite quotes out of context:

OpenWeedLocator Update

A step-by-step guided walkthrough of the OWL software installation is now available on YouTube. The video takes you through the two-line installation shown below.

git clone https://github.com/geezacoleman/OpenWeedLocator owl

bash owl/owl_setup.sh 

The owl_setup.sh script automates the process of updating the system, checking for camera connections, errors and virtual environment installation and setup. Those two lines should get your OWL ready to go. Besides software installation, I also go through features of the software itself and how to change configuration file settings to suit your needs. Check the video out below and let me know if there are things I have missed.

As a side note, the process of finding the best Raspberry Pi screen recorder (that works with Wayland…) was surprisingly challenging. After trying a number of the common ones, the command line option wf-recorder was super simple and effective on the latest versions of Raspbian using Wayland not X11. Works as suggested without fuss or too many issues, besides from running substantial updates during a few early takes.

In the pipeline

After some good progress by ElectroS3B on using the Raspberry Pi AI camera for weed detection, work is underway on integrating the Raspberry Pi AI Camera (with the IMX500 sensor) into the OWL software. You can see Sébastien’s work here with the IMX500 camera running his weed detection algorithm.

There is also work happening on supporting more advanced machine vision cameras from Lucid Vision Labs through their Arena SDK. This will enable cameras with higher quality sensors to be used if needed for data collection. The price point of these cameras is quite high, however, some are IP67 rated without additional enclosures.

Final Thoughts

In agriculture, we have an opportunity to create our own brand of open source development, one which centres farmer-driven innovation. Listen to many podcasts with CEOs, executives and founders of agtech business and one of the most commonly repeated ideas is that farmers drive their problem solving and solutions. Yet, a dominant business model in agtech is top-down—tools are developed in large companies and sold to farmers through dealer networks. Niche markets, problems and smaller growers become round holes buying square pegs. In the past, these pegs were steel and machinable, but now, black box AI and electronics-heavy equipment are leaving fewer options for growers to innovate.

By opening access to technology, like Eric Raymond's bazaar, we give farmers the chance to innovate organically, fostering a bottom-up approach.

Thanks again for tuning in every second week to read this newsletter, very much appreciated each time. Please keep the emails coming with questions and comments.

Until the next edition!

Cheers,

Guy