AI Assistant for Enigma2 Satellite Receivers

Finally got the break-through i was looking for with this project, and it should yield excellent results later this winter as time becomes available to use it.
It is now possible to have 2 AI's communicate and work directly with each other. In the past, you would have one AI to code, which was usually Claude Code. Something would be written --say the TNAP-wireguard vpn. Then it would crash when put the files into the receiver. Next, the AI that wrote the code starts listing commands to run on the receiver as debug. Files were then edited, transferred to the receiver and tested. The process repeats.... Now we have an AI on the Inside of the receiver that is root. This ai has direct access to what is happening in the receiver, and is also capable of editing files.

Where this puts things is almost in total automation. You now have one AI on the outside of the receiver in direct contact with the AI on the inside of the receiver. This will speed up development and debugging, probably by a very noticeable amount. Will be interesting to see. Granted, we always had the capability of the outside AI using ssh on the receiver, but having a real bona-fide ai inside the receiver changes things.

To go further, with the TNAP wireguard vpn, you could theoretically ssh into someone's receiver that is having problems, and send the ai's in to fix it. Right now, we have this crash log stuff, a file or fix has to be presented then tested...etc. What you can do with the TNAP tools available is pretty impressive --at least in theory. TNAP-wireguard only works for static ip addresses which not everyone has. However, you might get it to work with a non-static ip if the ip did not change during your connection.

Attached is a file of one ai, Claude Code testing the internal ai after making some adjustments to it. It is interesting for me to see 2 ai's working together on a FTA receiver. Training of the internal ai will continue as time permits....It's always something....This ai stuff is progressing at unbelievable speed.
 

Attachments

In July, 2025 this thread was started as a concept of sorts, installing the Ai into the fta receiver. Now in December 2025, I have the AI installed in the receiver, and producing some epg upon request. This whole affair will take probably a few more months to mature, but it is usable now for some things.

What I find that in a way that is hard to imagine is having one AI(Claude Code) supervising and instructing 2 Ai's embedded in different FTA receivers. This is rough and very much so an Alpha type project. But just the thought of having a setup like this is sort of mind blowing. You could just about start a factory with these guys!

Shown below is the Claude Code terminal window. Claude Code told one AI to do some tests, which succeeded, and at the same time was having the other Ai download and install an image into the receiver. The image install almost succeeded, or I guess they got halfway with it and it froze, bricking the receiver for a minute. Straightened out the receiver a bit later. The whole affair was quite comical to watch!

Question: Where we will be with this in 5 more months if work continues? Your guess is as good as mine...
Screenshot from 2025-12-08 01-27-17.webp
 
Jesus christ bro, next thing you will be inserting Claude up your bum to improve your efficency..
Do It Duck GIF by DefyTV
 
Current ai assistant version is 1.18.27-r0.
This project was almost scrapped several times due to an ai that did not know the job or could not do the job. A simple task for example: "Change the channel to <enter channel name here>" would result in the ai wandering about the receiver files, devouring resources called 'tools" or "tool usage" which costed an enormous amount of money. An unanswered question could cost as much as $1.10, depending on what the question was and what the resource usage was set at. It is one thing to spend time, but totally a different thing to spend time and money without seeing results!

The money situation grew even more when you consider that you are getting one ai to train another ai and have to pay both. But this is something I wanted and now have at a reasonable price. Then add the fact that our embedded ai is not trained by the developer for what we are doing, and you end up with the ai trying to execute commands that simply do not work in busybox-enigma2 Linux, but are common in ordinary Linux where the embedded ai has been trained. in other words, a Mess!

Also what I have experienced in training this ai is what others in the industry are experiencing in training ai's. You do not just drop the Command Line Interface or CLI in the fta receiver and automatically have an all-knowing ai assistant. Really the ai in our fta setup resides in the cloud, and not inside the receiver. Multiple ai types can be selected, but their cost varies. Originally I tried a low cost ai, but could not get it to work. Changing to a higher priced ai worked and was nice, until you asked a question or requested a command that was not fully understood, then the cash burn started.

The final solution for our TNAP ai is really a combination of things including hard-coded instructions, recipes for completing tasks, notes, and a detailed description of what enigma2 is and how it works. The detailed description for enigma2 was derived by looking at the entire block of enigma2 source code, then deriving how it functions or works. All of this information is stored so the ai can access it on start-up meaning we can immediately have an educated enigma2 ai instead of an educate idiot.

As solutions were found, it was realized that the higher priced ai was not needed, which ended up in a great operating cost reduction. Sure there may be cases where a higher priced ai is needed, but in reality enigma2 is not overly complicated, so a lower tiered ai can handle probably 90 percent of imagined tasks. The TNAP ai today or should we say the current version represents what I consider to be a useful product. For certain more debugging and training will be needed, but not many will use it anyway because of what it is.
 
Embedded ai used for crash log explanation

Fooling around and pressing the wrong remote buttons caused a crash. The ai correctly identifies the problem as shown below. Actual crash log is attached.
enigma2-crash 2025-12-11 16-42-11.webp
 

Attachments

Back
Top