What Lucid is, in one breath
Lucid is CongoSky's computational-knowledge engine — think of it as the honest opposite of a chatbot. A chatbot predicts plausible-sounding words. Lucid does the opposite: it runs the actual number when a question is a calculation, checks the fact against cited knowledge when it's a fact, and refuses when it can't stand behind the answer. The failure mode isn't a confident lie — it's an honest “I don't know.” And because it carries no giant model, it runs where models can't: offline, in your browser, even on a tiny chip in a $20 phone.
How it works — three minds, one honest answer
We don't build one giant know-it-all. We put three specialists together, each doing the one thing it's best at — with Lucid as the final gate that holds back anything unproven.
The brain · reasons
Claude Opus
The best reasoning we can get does the thinking. We rent this brain rather than own it, and you can bring your own key — Opus is simply the best one we point to by default.
The fact-checker · looks it up
Perplexity-style live search
When an answer needs a current fact, a live look-up fetches it with its source — so the claim arrives with a link you can follow, not a guess from memory.
The conscience · insists on correct
Lucid
Our own engine has the final say on what it holds: it computes, checks against its cited knowledge, and if it can't back something up, it refuses. The brake on the whole machine.
You ask → Opus reasons → live search checks the facts → Lucid insists it's correct or refuses → an answer you can trust
Honest detail: the brain is rented and swappable (bring-your-own-key), the live search is the fact-checker, and Lucid is the correctness gate on what we hold. We show you how the pieces fit — not the secret recipe that joins them.How Lucid is different from OpenAI (and the rest)
We use the same kind of powerful models everyone uses. The difference isn't the model — it's everything we built around it, and who it's for.
| On this | The usual AI | Lucid |
|---|---|---|
| When unsure | Confidently invents something that sounds right. | Says “I don't know,” or answers with a source. Refusing is a feature. |
| A number / a fact | Guesses from patterns in its training. | Computes it, or cites it. The answer is checkable. |
| Your private data | Flows into the model; may be kept or trained on. | Hidden from the brain before it reasons; the AI never sees who you are. |
| Where it runs | A giant datacentre you must reach over the internet. | In your browser, offline — even on a tiny chip with no signal. |
| Who owns the brain | One company owns the model and the lock-in. | Rented and swappable. Bring your own key. No giant to hold you hostage. |
| How you trust it | “Trust us.” | Every claim carries a re-runnable test you can check yourself. |
What you can do with it
Ask and get the truth
Real answers with their sources — or an honest “I don't know,” never a confident fabrication.
Speak your language
BuaBantu reads register and dialect — the meaning in the language people actually speak, not textbook flatness.
Works offline
No server, no account, no foreign cloud. It runs on the device in your hand, even in airplane mode.
Carry your context
Encrypted, compressed context that travels with you — small enough for the thinnest connection.
The honest scope — what's real today
The on-device engine is live: it computes, cites, and refuses, in your browser right now. The full “knows the world” version — ingesting the planet's useful knowledge at scale — is the road we're building, rung by rung, and we never wear a badge we haven't earned. What runs, runs; what's still a dream, we call a dream.
Check our working: the engine sits on a suite of re-runnable tests, all green, that run every time we change the code. See the proofs ↗