about
A small fire, kept close.
the idea
Most AI assistants are thin clients. You type, it ships your words to a datacenter, a model you can't inspect produces an answer, and a log of the whole exchange lives on someone else's disk indefinitely. That's a reasonable trade for some work and a terrible one for the rest — notes, contracts, medical questions, code you haven't shipped, anything you'd rather not hand over.
emb3r takes the other path. Open-weights models are now good enough to run on a laptop, so emb3r loads one into your own RAM and keeps it there. Slightly less capable than the largest hosted models, and completely yours.
questions
The things people ask before installing.
Is it really offline?
By default, yes, always. Downloading a model the first time needs a connection, because the weights have to come from somewhere, and emb3r checks GitHub on launch for a new version — that check sends nothing about you or what you've typed, just asks "what's the latest release?" You can turn your network off entirely afterward and keep chatting. There is no telemetry and no analytics, ever. The one opt-in exception is web access (see below) — off unless you turn it on yourself.
Is it as good as the big hosted models?
No, and it would be dishonest to claim otherwise. A small model on your laptop is not a frontier model on a rack of GPUs. It's very good at summarizing, drafting, answering questions about a file you attach, and everyday code help. It's weaker at long multi-step reasoning and obscure knowledge.
Will it slow my computer down?
While it's generating, it uses your GPU or a chunk of your CPU — you'll notice it on a thin laptop. Idle, it costs nothing but the RAM holding the model, and closing emb3r frees that right up.
What does it cost?
Nothing. Free and open source, no accounts, no paid tier, no seats. The only cost is disk space and electricity.
Which models can I run?
emb3r offers a curated set of open models — Llama, Qwen and Mistral, in a few sizes. It recommends one that fits your hardware, and you can download others and switch between them in Settings.
Does it remember past conversations?
Yes. Each profile keeps its own history, saved to disk and restored the next time you open emb3r — with real memory of what was said, not just old text sitting on screen. Switch between past chats or start a fresh one from the History panel.
Can I change how it talks?
Yes — the instructions that make Ember, Ember are editable in Settings, not hardcoded. Make it more formal, more terse, or something else entirely.
Where does my data go?
Nowhere, by default. Your messages go to the local model and no further. Settings, downloaded models and saved conversations live in a local folder on your machine that you can read or delete at any time. The only thing that can leave your machine is a question you've asked emb3r to look up online — see the next question.
Can it access the internet?
Only if you turn it on. Web access is off by default and needs your own free Gemini API key, entered in Settings. When it's on, emb3r decides per-question whether something needs current information (news, prices, "today") — and only that question, not your other chats or files, goes out to search the web for an answer. Turn it off any time and it's back to fully local.