How we build MeshSat
MeshSat is built with AI tools. We test it on real hardware, and we answer for all of it.
Meet the team and see the processWho builds it
We get asked about this a lot, so it’s all on one page.
MeshSat is a small team in Leiden, the Netherlands. One person directs the code: Kyriakos. Elli does the hardware assembly, and the verification and QA of everything. George is our PR and marketing consultant.
Fred
Our guest cat and team mascot.
How the code gets written
Most of the code is written by AI coding tools. Kyriakos decides what gets built, tells the tools what to write, and reads and tests what comes back.
We’re not the only ones working this way. In October 2026 Linus Torvalds published a Linux kernel branch whose code, test programs and review were done with an LLM coding assistant under his direction, with the measurements run on his own machine (his commit).
How it gets checked
- The Bridge, the Hub and both phone apps run automated tests on changes to their main branches. Required test failures block the build or release.
- The Bridge pipeline targets both of our field kits on every change to its main branch, the two prototypes in the photo below.
- A channel is only marked LIVE after it has run on the prototype hardware.
The numbers
Counted from the repositories on .
| First commit | 1 March 2026 |
|---|---|
| Days with at least one commit | 112 |
| Commits | 8,614 |
| Repositories | 14, of which 11 are public on GitHub |
| Lines of code | about 754,000 |
| Lines of tests | about 308,000 |
| Test functions | about 7,700 |
| AI tokens processed, 7 September to 21 October 2026 | about 81.6 billion |
| Of those, written by the AI tools | about 252 million |
That’s about four months of actual work. There were almost no commits between late April and late August.
How we counted
Non-blank lines in every programming language we use, tests counted apart. Docs, config, generated files and other people’s vendored code aren’t in it. In our fork of the Meshtastic firmware we only count the lines we added.
The tokens come from the session logs on our build server, where nearly all of the work happens. Logs from before 7 September 2026 are gone, so March to August isn’t in those two numbers. Most of the tokens processed are the tools reading code and context again that they’ve already seen.
Where each channel stands
Every channel carries one of five codes. They mean the same thing here, in the docs and in the Bridge README.
- LIVE
- Tested and working on our prototype hardware: the two V1 field kits, or a MeshSat node with a phone
- CODE
- In the Bridge, not running on a kit: either no kit carries the hardware, or we have no record of a run. It doesn't mean it failed.
- BENCH
- The hardware is on our bench, not in a kit yet
- COMING SOON
- The hardware is on its way to the bench. Nothing has been tested
- FUTURE PLAN
- Planned. Nothing is bought
The full table, with each channel’s current status and what each run showed, is on docs.meshsat.net.
MeshSat is a prototype under active development. It has never been deployed to a real user or used in a real emergency. It is not an emergency service.
What the tools don't do
- Pick the hardware, wire it or solder it.
- Take the kits out and run the tests on the radios.
- Answer for what we publish. We do.