How a Yaesu HT, a Raspberry Pi, and an AI agent ended up sharing a digital voice channel, and with me.
I key up a Yaesu FT-70D handheld and ask, out loud, "Bud, is the shed leaking power right now?" A few seconds later, the radio answers back, in a voice, over the air, the way a person would. That's Bud-RF: a voice assistant that lives on a digital ham radio channel instead of a phone or a smart speaker.
Bud-RF is a way to talk to my home computer system using nothing but a radio and a PTT (push-to-talk) button. No app, no screen, no typing. I key up on my HT, ask a question or give it something to remember, un-key, and it answers back out loud through the same radio, on the same digital voice channel it heard me on.
It's built on top of "Bud," the AI agent that already runs my home lab and keeps notes on everything from the weather station on the roof to the solar panels out back. Bud-RF is simply an RF doorway into that same agent: it rides on Yaesu's System Fusion (YSF) digital mode instead of Wi-Fi. Whether I reach it by radio, by text message, or from a laptop, it's the same brain answering.
Think of it like a walkie-talkie version of asking a smart speaker a question, except instead of Wi-Fi, the question travels over the radio waves amateur radio operators (hams) have used for a century, and instead of a generic assistant, it's mine, trained on my own house and projects.
None of this needed exotic hardware. The whole thing is a relay race: each stage decodes one protocol and hands the audio to the next, from RF all the way down to plain PCM and back up again.
The round trip, my voice going out and Bud's voice coming back, takes just a few seconds. Under the hood it's a chain of small, mostly off-the-shelf pieces glued together with some custom code: a low-cost YSF hotspot the size of a pack of cards running MMDVMHost, a Raspberry Pi acting as the bridge, an AMBE vocoder decoding the digital voice codec down to raw audio, speech-to-text (STT) turning that into words Bud can reason over, and text-to-speech (TTS) turning Bud's reply back into audio that gets re-encoded and injected as a proper YSF frame, not just played over a speaker into a microphone.
I built it by getting each hop in that chain proven on its own before wiring it to the next: confirm the hotspot decodes a live burst correctly, confirm the vocoder's bit-mapping produces clean audio both directions, connect the agent, then confirm a reply survives the full round trip back out over RF without getting mangled. Most of the debugging turned out to be a port-configuration mismatch between the audio bridge stages, not a codec problem, the kind of thing that only shows up once you've got real traffic flowing through every link at once.
"RF" just means radio frequency, the actual radio signal, as opposed to Wi-Fi or the internet. "YSF" is Yaesu System Fusion, the digital voice mode my radios use; a "hotspot" is a small low-power device that bridges a handheld radio's RF signal to a network, the way a Wi-Fi router bridges a laptop to the internet. When you see "Bud-RF" in these notes, it means "the version of Bud you reach over the radio," to distinguish it from the version you'd text or type to.
Bud-RF isn't a novelty that can only say "hello." It has real jobs. Here's a sample of what a conversation over the radio sounds like:
Ask for the status of the home systems, cameras, and sensors while I'm away from a screen.
"Bud, is everything running okay at home?"Pulls from my own rooftop weather station, not a generic forecast, plus live radar and severe weather outlooks.
"Bud, what's the wind doing right now?"Battery charge, voltage, and how much sun the solar setup is pulling in at that moment.
"Bud, how's the solar battery looking?"Knows my rough location from my phone and can answer road, traffic, and travel-time questions while driving.
"Bud, how long to get home from here?"Speak a reminder or a to-do out loud and it's waiting for me later, without ever touching a phone.
"Bud, remind me to check the antenna mount."Ask it how to work a feature on one of my radios and it reads back the instructions, hands-free.
"Bud, how do I set up cross-band repeat?"A quick sweep of every service and hotspot in the home lab: which ones are up, which need a look.
"Bud, is CyberFusion linked right now?"Bud isn't the only AI working on this home lab. I can speak a task straight into a shared ticketing system, and one of the other agents on the project picks it up next.
"Bud, open a ticket to get the solar dashboard's charging graph fixed."It also remembers the conversation. If I ask a follow-up a minute later, it knows what "it" refers to, the same way a person would, rather than treating every question like the first one it's ever heard.
"Second Brain" is what I call my personal notebook of everything I've learned and documented about my home lab, my radios, and my projects: years of notes, kept in one place, that I can search instead of relying on memory alone.
Bud-RF can draw on that notebook to answer questions accurately, instead of guessing. If I ask how a particular radio's split-frequency mode works, it isn't improvising. It's reading from notes I wrote (or asked Bud to write) after actually figuring it out the first time.
This matters more than it might sound like at first. A generic assistant can tell you generic facts about ham radio. Mine can tell you specifically how my setup works, because it's reading my own documentation, not guessing from what's common on the internet.
The connection runs one careful direction: Bud-RF can read curated, radio-safe notes from the Second Brain to answer better. What I say back to it over the air, a passing comment, a location, a stray thought, goes into its own separate notes, not straight into the permanent notebook. That's on purpose: radio conversations are public by nature (anyone with a receiver can technically hear them), so nothing sensitive is exposed, and nothing gets permanently filed away without a second look first.
If you're not familiar with amateur ("ham") radio: it's a century-old hobby built around a simple idea. Licensed operators, each issued a callsign (mine is WY6Y), can use dedicated radio frequencies to talk to each other, sometimes across the room, sometimes across the world, with no phone company or internet provider in between. Ham radio still matters today because it keeps working when everything else doesn't: after storms, power outages, or when cell towers are overloaded.
Modern ham radio also has digital voice modes. System Fusion is Yaesu's version, and there are rival flavors from other manufacturers (DMR, D-STAR) that work on the same basic principle, carrying voice as compressed digital data instead of a raw analog waveform. That's what makes something like Bud-RF possible in the first place: the signal can be handed off to a computer as cleanly as a VoIP call would be, without needing the internet to reach the radio itself.
Building an AI agent that speaks that language felt like a natural next step: not replacing the radio hobby with technology, but bringing the two together.