🌸 LLMs at the Edge of Connectivity
I built Local LLM Server because I thought it was a cool idea. The app exposes Apple Intelligence’s on-device model as an API, compatible with both Ollama and OpenAI formats. You can point any local chat app at your iPhone and query an LLM without touching the internet.
My original thinking was that students or researchers might want to run unlimited queries against an LLM without worrying about API costs. Since it runs locally, the only cost is electricity. But I never had a clear picture of who would actually use this for extended offline work.
Then I met Nik on Twitter.
A Real Use Case Emerges
Nik is a communications engineer who works with RF systems. He reached out asking about Local LLM Server, and when I asked what he had in mind, his response surprised me:
I'm working sometimes in isolated environments from the internet. It can give me real coding assistance with a development workflow on device.
— Nik (@c_sharp_gr) February 17, 2026
He elaborated on the constraints:
In non-urban areas tablets are bulky, laptops even more so if you need to work one-handed and make it useful not only for coding. I had situations with no internet over UTP and the use case becomes real. Just uncommon.
— Nik (@c_sharp_gr) February 17, 2026
I had imagined this scenario but never expected to actually encounter it. Someone who needs to write and debug code, but is physically beyond the reach of the internet. Not on an airplane for a few hours, but regularly working in connectivity deserts where the internet simply does not exist.
An on-device LLM becomes genuinely valuable in these places. Not just for coding assistance, but for reference material: RF standards, path loss formulas, antenna gain calculations, spectrum allocation tables. Information that would normally require a quick search can instead be queried from your pocket.
Connectivity Deserts
Nik’s situation is specific, but the pattern is not. There are entire industries where people do technical work beyond the reach of the internet.
Offshore oil rigs are an obvious one. Engineers maintaining drilling equipment and industrial control systems work on platforms far from shore. Satellite internet exists but can be unreliable, expensive, and bandwidth-limited. Troubleshooting PLC code or referencing mechanical specifications often means waiting hours for a connection that may not come.
Remote research stations are another. Antarctic bases, rainforest field stations, mountain observatories. Scientists at these locations work with specialized equipment that might malfunction at inconvenient times, and their connectivity ranges from spotty to nonexistent.
Then there is disaster response. When infrastructure fails after earthquakes, hurricanes, or floods, first responders and engineers work in areas where cell towers are down and landlines are destroyed. Having technical reference material and coding assistance available offline could help restore critical systems faster.
This Was Not Possible Two Years Ago
What strikes me most about this is how recent it all is. Ollama launched in August 2023, making it possible to run LLMs locally on a laptop. Apple Intelligence followed in October 2024, bringing on-device models to the iPhone. Before that, useful LLMs required cloud connectivity.
Now Apple has baked a capable model into the operating system, and apps like Local LLM Server can expose it to any tool that speaks a standard API.
Apple Intelligence is nowhere near frontier models. But it does have a general world model with useful reference knowledge baked in. Someone on a submarine or oil rig can now query that knowledge without any infrastructure at all. The capability travels with you.
If You Work at the Edge
I built Local LLM Server as a fun experiment. Nik showed me it solves a real problem in places I had never thought about. There are people doing serious technical work in connectivity deserts all over the world, and for the first time, they can carry an LLM in their pocket.
If you work in one of these environments, I would love to hear what you would use an offline LLM for. You can find me on Twitter @_KevinTang or try Local LLM Server yourself.
I built [Local LLM Server](https://apps.apple.com/us/app/local-llm-server/id6757007308) because I thought it was a cool idea. The app exposes Apple Intelligence's on-device model as an API, compatible with both Ollama and OpenAI formats. You can point any local chat app at your iPhone and query an LLM without touching the internet.
My original thinking was that students or researchers might want to run unlimited queries against an LLM without worrying about API costs. Since it runs locally, the only cost is electricity. But I never had a clear picture of who would actually use this for extended offline work.
Then I met Nik on Twitter.
## A Real Use Case Emerges
Nik is a communications engineer who works with RF systems. He reached out asking about Local LLM Server, and when I asked what he had in mind, his response surprised me:
<blockquote class="twitter-tweet"><p lang="en" dir="ltr">I'm working sometimes in isolated environments from the internet. It can give me real coding assistance with a development workflow on device.</p>— Nik (@c_sharp_gr) <a href="https://twitter.com/c_sharp_gr/status/2023635613111513530?ref_src=twsrc%5Etfw">February 17, 2026</a></blockquote> <script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>
He elaborated on the constraints:
<blockquote class="twitter-tweet"><p lang="en" dir="ltr">In non-urban areas tablets are bulky, laptops even more so if you need to work one-handed and make it useful not only for coding. I had situations with no internet over UTP and the use case becomes real. Just uncommon.</p>— Nik (@c_sharp_gr) <a href="https://twitter.com/c_sharp_gr/status/2023673323847028806?ref_src=twsrc%5Etfw">February 17, 2026</a></blockquote> <script async src="https://platform.twitter.com/widgets.js" charset="utf-8"></script>
I had imagined this scenario but never expected to actually encounter it. Someone who needs to write and debug code, but is physically beyond the reach of the internet. Not on an airplane for a few hours, but regularly working in connectivity deserts where the internet simply does not exist.
An on-device LLM becomes genuinely valuable in these places. Not just for coding assistance, but for reference material: RF standards, path loss formulas, antenna gain calculations, spectrum allocation tables. Information that would normally require a quick search can instead be queried from your pocket.
## Connectivity Deserts
Nik's situation is specific, but the pattern is not. There are entire industries where people do technical work beyond the reach of the internet.
Offshore oil rigs are an obvious one. Engineers maintaining drilling equipment and industrial control systems work on platforms far from shore. Satellite internet exists but can be unreliable, expensive, and bandwidth-limited. Troubleshooting PLC code or referencing mechanical specifications often means waiting hours for a connection that may not come.
Remote research stations are another. Antarctic bases, rainforest field stations, mountain observatories. Scientists at these locations work with specialized equipment that might malfunction at inconvenient times, and their connectivity ranges from spotty to nonexistent.
Then there is disaster response. When infrastructure fails after earthquakes, hurricanes, or floods, first responders and engineers work in areas where cell towers are down and landlines are destroyed. Having technical reference material and coding assistance available offline could help restore critical systems faster.
## This Was Not Possible Two Years Ago
What strikes me most about this is how recent it all is. Ollama launched in August 2023, making it possible to run LLMs locally on a laptop. Apple Intelligence followed in October 2024, bringing on-device models to the iPhone. Before that, useful LLMs required cloud connectivity.
Now Apple has baked a capable model into the operating system, and apps like Local LLM Server can expose it to any tool that speaks a standard API.
Apple Intelligence is nowhere near frontier models. But it does have a general world model with useful reference knowledge baked in. Someone on a submarine or oil rig can now query that knowledge without any infrastructure at all. The capability travels with you.
## If You Work at the Edge
I built Local LLM Server as a fun experiment. Nik showed me it solves a real problem in places I had never thought about. There are people doing serious technical work in connectivity deserts all over the world, and for the first time, they can carry an LLM in their pocket.
If you work in one of these environments, I would love to hear what you would use an offline LLM for. You can find me on Twitter [@_KevinTang](https://x.com/_KevinTang) or try [Local LLM Server](https://apps.apple.com/us/app/local-llm-server/id6757007308) yourself.