AiNavigatorSAYWorking now
A suggested response
A natural sentence or two you could say next. Editable words for you — never a reply the AI delivers on your behalf.
The principle
Not a chatbot with a microphone.
The output isn’t a transcript, and it isn’t an essay. It’s a handful of short cards you can glance at mid-sentence — written in your voice, for you to adapt and say yourself.
ASKWorking now
A better next question
A follow-up that moves the conversation forward, or uncovers what the other person actually cares about.
KNOWPlanned
What you already know
Relevant facts from your own documents — CV, projects, notes — used only when you ask for owner-tailored answers, with sources.
WEBPlanned
What’s happening now
Current information on the topic at hand, found with a minimal search query — never by sending the whole conversation.
How it works
Start once. Then just talk.
Live mode is the default: one tap starts a visible session, and the cards keep pace with the conversation. No tapping for every question, no switching screens.
- Microphoneyou start it
- Speech to texton the phone
- Conversation enginerecent context, intent
- Your chosen modellocal or your gateway
- SAY · ASK cardsfor you to use
Start live
Tap Start live once. The microphone indicator and session notice make it obvious that the assistant is listening.
Speech becomes text
Audio is transcribed by the phone’s on-device speech engine, in memory. Raw audio is never saved or sent to a model.
Cards keep up
Each meaningful, finished turn is interpreted with a small window of recent context. New questions replace stale ones; repeats are suppressed.
Pause or end anytime
Pause, End or leaving the app stops capture and cancels pending work. Nothing restarts until you say so.
Prefer more control? Manual mode listens until you tap Stop & interpret — the microphone switches off before any AI reasoning begins. You can also type what the other person said, without using the microphone at all.
Your settings
You choose how it thinks.
Every choice is explicit, visible and yours to change. Nothing switches behind your back — not the model, not what gets shared.
Where reasoning runs
On-deviceYour gatewayDemo
Use a model on the phone, or connect your own OpenAI-compatible HTTPS gateway — an NVIDIA Nemotron preset is included. It never silently falls back from one to the other.
General or about you
GeneralOwner-tailored
General answers the question on its merits. Owner-tailored adapts to your verified background — and if the evidence isn’t there, the card says so instead of inventing experience.
The tone of voice
NeutralFormalFriendlyHumorousConciseCustom
Tone shapes the wording, never the facts. Humour stays respectful, and a custom style is a short preference — not a new set of rules.
“Silence is better than a confidently wrong suggestion.”
When there isn’t enough context or evidence, AiNavigator shows nothing rather than guessing. Model output is validated before it ever reaches a card, and unsupported personal claims are dropped.
Private by design — and honest about the limits
An assistant that listens has to earn trust. AiNavigator is built around explicit control, not convenience.
You start it, you see it
Capture is always started by you, stays in the foreground and is clearly visible. No background, hidden or locked-screen listening.
No raw audio leaves
Audio is processed in memory by the on-device speech engine. It is never stored as a file and never sent to a model.
Bounded, approved text only
With a gateway, only a short slice of recent finished text is sent — after you approve that exact destination and model. Live mode needs a second, separate approval.
Keys stay locked
API keys are encrypted with Android Keystore, bound to their destination and excluded from backups. They never appear in prompts or logs.
Nothing kept by default
Transcripts and cards live in memory for the session. There are no app analytics, and logs contain timings, not content.
The fine print
The phone’s speech service may use the network for setup and its own telemetry, so “on-device” is not a promise of zero network activity. Text already sent to a provider can’t be recalled.
Use AiNavigator only in conversations where everyone involved is comfortable with audio being processed. Consent requirements vary by place.
Where it is today
A working prototype, built in honest slices.
Each milestone keeps the app runnable on a real phone. Here’s what’s real now, and what comes next.
Working now
- Hands-free Live mode, plus Manual capture and Stop & interpret
- On-device speech-to-text (English, US)
- Real SAY and ASK answers through your own HTTPS gateway
- General or owner-tailored answers, with selectable tone
- Keystore-protected keys and per-destination sharing approval
- An offline Demo mode with fixed example cards
Coming next
- On-device LLM reasoning — Nemotron E4B is a candidate, not yet verified on a phone
- Personal knowledge: import your own documents for sourced KNOW cards
- Web research: WEB cards with links to their sources
- Latency, battery and long-session measurements on real devices
- Later: speaker awareness and a glanceable overlay
Early access
Want to try it in your own conversations?
AiNavigator is a prototype, and early testers shape what it becomes. Request an invitation and we’ll get in touch when a testing place is available for your phone.
- Built first for Galaxy S23 Ultra and Galaxy S25 Edge. Other Android 12+ phones may work, but aren’t tested yet.
- Real answers currently come through your own AI gateway and key; an offline Demo mode needs neither.
- Testing invitations and AiNavigator early-access updates only.
Prefer email? Write to ainav@kuara.dev.