Do PFC's work as intended?


My attention was first brought to the Open Agriculture Initiative sometime in 2016 when I watched Caleb’s TED-Talk. The idea of replicating desired results in climate controlled growing chambers based on shared recipes fascinated me. Just as the vision of a large community of PFC owners did that were continuously perfecting their results and sharing recipes and insights. I dreamt of growing my own tomatoes that tasted just as the ones bought at a market in southern Italy. I wanted to be part of this and started thinking about building and operating a PFC – and later a Food Container.

I started collecting information and talked to everybody about this idea, looking around trying to collect more information on the state of the project. My enthusiasm started to drop when I seemed to recognize that no actual climate recipes were available that could be reliably used to produce results. My dream of perfect tomatoes waned.

I also thought of the problem of seeds. How to make sure I have the right seeds to attain the desired results? And the problem of resetting a growing chamber to the exact same condition when starting a new recipe. What I if any contamination from a prior grow phase is still lingering and influences the results? And the aspect of other plant and soil interaction that surely influences results considerably and would be difficult to adjust to.

Most of all the chatter in the forum pointed the many groups trying to build PFC’s but very little in actual PFC’s fully automatically growing stuff based on climate recipes.

This brings me back to my question: Do PFC’s actually work as I understand them? Do they work completely autonomously? Do they record their recipes and upload these to a repository in the cloud? Are these accompanied by information describing what the recipe actually grows and what qualities are achieved? Are the recipes continuously improving? Can I expect to reliably go through many grow cycles without major overhaul every time? Is a scenario realistic, where multiple PFC’s in my basement are chugging along producing good food – with little or no supervision?

If not: what seems to be missing? I am in a situation where I could invest considerable effort into building and perfecting a PFC. But I would like to be able to assess my chances before I dive in. Can anybody gibe me information on all my questions?


It depends. How do you understand them? Do you expect PFCs to meet the standards for an established commercial product from a startup company? In that case, you will be disappointed. If you want to feed yourself with food grown in current PFCs, you will be disappointed [the main issues are cost effectiveness, reliability, and small crop size].

On the other hand, do you think of the existing PFC designs as prototype devices that fit into the long term vision for a research project which might take many years to accomplish its goals? In that case, your expectations are right on target. For the questions you asked about features, cloud connectivity, and recipes, the current answer (pfc-edu v3.0) seems to be, mostly yes, but just barely, on a limited basis, and with many qualifications. There are good reasons to expect that the situation will improve, but progress will be slow. That’s normal for early to mid-stage prototypes from a small research team.

Also, PFC prototypes have been successfully used in schools where the goals are to teach kids about plants and inspire them to learn about science and technology. If you think of PFCs and Caleb’s talks as a source of inspiration for you to learn more about indoor farming, they are already a great success in that regard.


Caleb is great as a promoter of new ideas, and casting a broad vision. The execution of that vision is more complicated. I believe it is possible to reproduce the conditions for growing a tomato that tastes like one from a market in southern Italy, but I don’t think we have a) the equipment to get the level of control over an environment, or even b) know all the factors that contribute to that flavor. My LED light is not a reproduction of the Italian sun.
I believe we will achieve those conditions, but it may be another 10 or 20 years before it is practical. These are hard problems. Just getting agreement on terms and standard data collection for consistent analytics is near impossible at this time.
Yes, PFCs work. I have lettuce and herbs growing within arms reach of me as I write this. Have they lived up to the promises? That is harder to answer


Thank you for these answers! Congratulations on a forum that really works. Many new questions come to my mind…. sorry for the long rambling ;-(

I can appreciate that I can’t feed myself with what can be grown in a single PFC – especially v3, which seems to have a considerably smaller grow space.

What I still can’t grasp: it seems to me that building a PFC and getting it to run is the major hurdle at the time. The whole issue of recording recipes, empirically improving and sharing them in a common repository doesn’t seem to be in the foreground – or then I haven’t found it.

What I understand from your answer: I cannot expect to choose from a myriad of climate recipes available in a cloud repository, download a chosen recipe into my PFC, pop in the seed(s) and leave the PFC more or less unattended until I can harvest.

Can I record my own growing conditions into a recipe and adapt/change/improve these for a second batch – e.g. raise temperatures half a degree, add more X or Y. Then check out the results, rinse and repeat? Thus by pure trial and error improving my recipe?

I have the impression that actually building a PFC with complete automation (all parameters are controlled by a recipe and automatically set by the PFC) is quite challenging – enough so that hardly any fully automated units exist – if any at all? If fully automated PFC’s exist: what are they being used for? I had the idea that fiddling around with grow parameters and perfecting results/recipes would be the main task. Then it should be possible for anyone to replicate these results. Or even better: transfer the recipe to a larger growing unit and producing on a larger scale. How many cycles can a current PFC be used for with constant results? What kind of issues must be dealt with from crop to crop in order to guarantee the same quality of results (pasteurizing the growing chamber?)?

I understand that one of the objects is to record not only the recipe, but also the results and then let some kind of AI make suggestions as to how to change input parameters to improve the process – getting away from trial and error. From what I understand the standards for data collection and analytics have not been established yet. Which gets me back to trial and error. Trial and error sounds OK - if we can get to a large number of PFC’s and farmers trying it out all over the world. I have my doubts if we can reach this situation – the cost of building and running the PFC’s seems to high. It seems that the individual PFC’s are not standardized enough and many builder have major issues getting the parts, assembling and getting it to run?

What I understood from Caleb’s presentation is that current produce in chain grocery stores is optimized for looks, shelf life, ease of transportation, resistance to herbicides & pesticides – taste and nutritional value seem to be at the end of a list of priorities. If I can grow my vegis in my own garage, many of these issues vanish. I can concentrate on what counts for me – taste and nutritional value.

Coming back to the example of tomatoes fresh from a market in southern Italy. I am quite sure that taste depends on the seed and soil. I have experienced tomatoes bought at a local market in Bulgaria and tomatoes grown of the same seeds here in a greenhouse in Switzerland – the taste made the journey and was more or less the same grown here. As with salmon or flamingos, who get their color from the food they eat: I’m quite sure taste depends to a large extent on what nutrients go into the mix (other things being at least acceptable).

So if I get the right seeds and have a growing chamber that I can set to replicate ideal growing conditions, I can get good taste and value – simply by fiddling with nutrients, light, temperature, etc. I should be able to improve the result to an optimum. The task of perfecting the nutrients with all the trace elements found in a good soil – or at least the ones that influence taste – could be quite damming.

It comes down to the question: is a PFC a growing chamber which can reliably and automatically replicate a recipe for many cycles and is it big enough to allow growing large plants like tomatoes? If this is the case I am willing to give it a try – maybe build a PFC, improve it, then build five and try perfecting results along the lines I describe.

What surprises me is your mentioning that LED lights cannot be compared to sunlight. From what I understand these lamps can be optimized for what spectrum of light plants thrive on? It seems to me that the Dutch indoor farming industry must have solved this problem?


You have identified a key issue.

Personal Food Computer is a marketing term for a specific new approach to an old idea. Researchers have been using plant growth chambers for many years. If you search for NCERA-101 and the Growth Chamber Handbook, you can read about some of the existing standards for data collection.

The difficulty is that existing growth chambers are large, very expensive, and not built according to Open Source principles. In order to apply machine learning to growth chamber results, you need many growth chambers. As I understand it, this is why Caleb came up with the PFC concept–to develop significantly less expensive growth chambers that will enable large, distributed, citizen science data collection experiments. Once a lot of data is collected, it will be possible to apply machine learning algorithms to the data set and identify patterns that can be used to make recipes. Once recipes are developed, then they could be applied in controlled environments that are designed for growing crops with a focus on cost effectiveness and yield.

It’s important to understand that Caleb’s vision for the future has not been accomplished yet, and if it happens, it will take a while. It’s probably more reasonable to look for progress over the course of years rather than weeks or months.


Good morning,

I read your discussion with a lot of interest…Should I understand that version 3.0 of the PFC does not offer for now a fully automatic recipe control ( given that a recipe would exist) ? Are the optionial sensors only meant to read the growing parameters ( EC, PH, humidity) but not meant at this stage to feed a climate control loop ? From what I’m reading, it seems that version 2.0 offered an automated controlled environment.

Thanks in advance


It depends on how you define “fully automatic recipe”. The PFC 3.0 design is able to actuate light spectrum changes over time. If you put several of them in a climate controlled room to maintain steady temperature and humidity, it would be possible to run experiments to measure and compare the response of plants to different light recipes. The pH and EC sensors could be used for human-in-the-loop experiments with nutrient solutions.

You can look at the OpenAg team’s PFC 3.0 recipes here:


Let me give three examples to illustrate the complexity of a controlled environment .

  1. LEDs are not the sun. Yes, we can cover the full spectrum, but I don’t get a sun-burn looking into my box, or worry that I will damage my eyes looking into the lights. At this time we are not trying to put small thermo-nuclear generators in the box, but create a light environment that gets us decent growth. It is a stretch to say that at this time we are recreating an outdoor environment in a box.

  2. Right now the MVP manages temperature by exhausting the warm air. That controls the temperature, but at the same time the exhaust air (and air inflow) change the CO2 and humidity (in a less controlled manner). It is a complex problem to control temperature without air exchange. Then think about CO2 or humidity control in a sealed environment (on a submarine or in space). That level of control becomes complex (and expensive) very quickly, especially at a small scale.

  3. It is easy to write a recipe and record data. The MVP, OpenAg, Fairchild Gardens (NASA) and ESA are all doing it - and we all are following different ‘standards’ and have been unwilling to work together on a common data standard. We are not trying to be perverse, but we have different goals and ideas. Think about defining light. You start with spectrum, duration and intensity. then you want to add a day/night cycle, then someone wants to have a progressive spectrum shift (reproduce sunset), or vary the intensity during flowering verses other growth stages; or have separate IR lights they can ‘flash’ just before turning off the other lights during flowering. How do you design a data structure that is flexible enough to capture all of these variable combinations? What boundaries/limits do you set; and how do you get several divergent groups to agree on those limits?
    My intention here is not to be discouraging, but to show how complex this problem is; and how amazing it is that we have come this far this quickly.

"Backend" openag-cloud-v1 Q&A for PFC 3.0

I was trying to find a end-user manual ( or a few pages) describing what to expect and how to use the current version with its current limitations. Does this exist ?


From what I’m reading in the many conversations, am I correct in crossing out the items in the here below diagrams to describe the current state of development? I’m crossing out actuators as I think it should be just just a light intensity and spectrum actuator. Please correct me if I’m wrong. We want to build 5 of those PFCs but we would like to know for sure the current functionalities.

  1. For the cloud UI part, please take a look at the links in this comment from the PFC_EDU announcement thread: Meet "PFC_EDU," the new PFC v3.0.

  2. For the software that runs locally on the PFC_EDU, you should probably read all of the .md documentation in

I think you’ve misunderstood the current situation. If you haven’t studied the code carefully yourself, or if you don’t have a senior developer on your team who is qualified to do it for you, I’d recommend that you put your hardware manufacturing plans on hold until you solve that problem. I phrase my suggestion so strongly because it will cost you a lot of time and money to build 5 prototypes. But, you won’t be able to use them without getting the software working. To use your hardware, you will need to be prepared to track OpenAg’s software changes as they potentially diverge from the hardware version you build. You may also want to change some of the current functionality before using it. For example, see issue number 135 in the openag-device-software repo and its corresponding commit, 260200b.

Also, please read this post from Rob earlier today about the current cloud stuff: "Backend" openag-cloud-v1 Q&A for PFC 3.0