Silas / by RMTE AI
The backcountry assistant that works with no signal.
Silas runs a language model on the phone in your pocket. Ask what those clouds mean, how to treat a rolled ankle, or for a story around the fire. When the answer matters, it comes from a reviewed source. When it does not have one, it says so.
How it answers
With no signal, the model and the knowledge pack both live inside the app. A question goes in, the pack is searched, and an answer comes back, all on the device.

You
What should I look for in these clouds?
Silas
Flat dark bases with tall tops mean a storm is building. Get below the treeline before it arrives.
The app
- 01Your question goes to the app on the phone.
- 02The question is turned into a vector and the pack is searched for the closest passages.
- 03The matching passages come back to the model.
- 04The language model writes an answer and labels where it came from.
01 / The problem
You can download a map. You cannot download an answer.
Out there the questions keep coming, and the ways to answer them run out. So the answers have to come with you.
The questions do not stop.
Is that cloud a problem. What is this plant. How far to water. Who was the guy this pass is named after. Off the grid you ask as much as you ever do. Probably more.
The answers stop at the trailhead.
Past the last bar, what you know is what you carried in your head, and what the people around you happen to know. That is the whole library.
You cannot download the unexpected.
You can save the route, the maps, the photos, a field guide or two. You cannot save the answer to a question you have not thought of yet, and those are most of them.
So the answers have to come with you. Silas stands on two things.
What we stand on
It works at zero bars.
Every core feature is tested in airplane mode. If it only works online, it is an add-on, not the product.
What we stand on
Honest about where an answer came from.
Every answer says whether it came from a reviewed pack or from the model's own knowledge. On safety topics, a guess is never presented as fact.
02 / Two roles
Trusted when it matters. Company the rest of the time.
It is always clear which role it is playing. The second is what makes people open it every evening. The first is what makes them trust it.

Trusted source
Weather hazards, wildlife, plants, first aid.
Answers come from reviewed knowledge packs and cite their source. If the packs do not cover it, Silas says so rather than guessing.
- What should I look for in these clouds building over the ridge?
- Is this water hemlock or cow parsnip?
- How do I treat a rolled ankle two days from a road?

Trail companion
Everything else.
Why there are 365 days in a year. How beavers shaped America. A ghost story for the campfire, or a short story about whatever someone is bored enough to ask for. These come from the model’s own knowledge and are labeled that way.
- Why are there 365 days in a year?
- How did beavers shape America?
- Tell me a story for the campfire.
03 / The whole trip
One assistant, from the trailhead back to the truck.
Ask it the way the moment allows, carry the knowledge the trip needs, and let it catch up when the signal comes back.
Ask by photo
Point the camera at a plant, a cloud bank, or a set of tracks and ask about it.

Ask by voice
Say the question and hear the answer. For rain, cold, and hands that are busy.

Curated knowledge packs
Reviewed knowledge for the things that matter out there: weather, wildlife, plants, first aid. Every passage traced to its source.

Build your own packs
Your notes, a guidebook, the plan for next week's trip. Hand them over and they become a pack you can ask, with no signal.

Better when you are back in range
Your trail conversations sync, an offline answer can be checked against a bigger model, and the next trip gets planned with one.

Everything above works with no signal. The last one uses a connection when there is one
The world does not stop when the signal disappears. Neither should your AI.
Waitlist
Be there when the signal drops.
Early access and occasional notes on what we are building. No noise.