Your AI sidekick for everything you say out loud
Point it at a meeting, a conversation, or nothing in particular. SID3KICK records, transcribes, summarizes, pulls out the calendar events and to-dos you mentioned, and rolls the whole day into a digest you can actually search later. On your device by default — the offline engine never sends a byte anywhere.
$5/month at launch · plus your own AI API usage · $0 if you run it fully offline
Tap to record when you know you want a transcript. Or leave it running and let it work out for itself when someone's actually talking.
Start recording and the pipeline takes over — transcribe, summarize, title, and scan for anything that sounds like a date or an appointment. An optional floating bubble sits over other apps so you never have to open SID3KICK to hit record.
A foreground service listens with the screen off and the phone in your pocket. Voice activity detection slices continuous audio into real speech, so you're not paying to transcribe silence — and each chunk runs the same pipeline.
It transcribes what's actually being spoken, not what your phone's locale expects. Then pin one output language and every title, summary, and digest comes back in it — which also inoculates you against cloud models hallucinating a wrong-language transcript on noisy audio.
Dates, times, and appointments get pulled out of what you said and surfaced as editable one-tap calendar suggestions. They write straight to your device's own calendar — no Google Cloud project, no OAuth screen, no API key.
"What did I say I'd send Priya?" "When was that dentist thing?" The Ask tab answers across every transcript at once, and because each capture is tagged with when — and optionally where — it happened, "yesterday" resolves against your real history instead of being guessed at.
To-dos get pulled out of what you actually said, land on a tickable list, show up in the daily digest, and — if you paste a token — push one-way to your Google Tasks default list.
Location tagging is off, country-only, or precise — your call. Country-only is the default: it reads the last-known fix with no GPS wake-up, keeps the place name, and blurs the coordinates to roughly a kilometre before storing anything.
Keep transcripts, summaries, to-dos, and digests in step across your devices through your Dropbox, Google Drive, or FTP server — never through anything we run. Each device writes its own snapshot and merges the others.
Dark and Light are ordinary Material 3. LCARS is not: black panels, asymmetric corners, condensed uppercase type, and synthesized interaction beeps generated at runtime — no ripped audio, no licensing headache.
Every stage is either on your phone or on a provider you hold the key to. There is no middle tier operated by us.
16 kHz mono audio, manual or voice-activated. Snap photos or short video alongside it — with video, only the audio is ever extracted and sent; the file itself never leaves the device.
Vosk on-device across ~20 downloadable language models, with zero network calls and zero cost. Or route to OpenAI's cloud transcription when you want the accuracy and don't mind the upload.
Title, summary, extracted events, action items, and optional speaker diarization — through OpenAI, Anthropic, a local Ollama or LM Studio box, or any endpoint that speaks the OpenAI chat-completions schema.
Day-by-day digests, a map of where you were, describe-and-recall search for half-remembered conversations, and a chat box that answers across the lot.
Plenty of apps say "privacy-first" and mean "we encrypt it on the way to our servers." SID3KICK has a configuration where there are no servers.
On-device Vosk for speech-to-text plus a local Ollama or LM Studio box for summaries, and nothing leaves your network. Not a degraded fallback — a first-class setup.
When you do use the cloud, you paste your own API key and talk to that provider directly. We are not a proxy, we never see the traffic, and there is no account to create.
Sync runs through storage you already own. Raw audio, photos, and video stay on the device that captured them; optional cloud archive and auto-prune only ever delete audio that's already been backed up.
Two separate bills, and we're upfront about both. The subscription is ours. The AI usage is billed by whichever provider you point the app at — or it's nothing at all, because you're running the models on your own hardware.
The app
$5 / month
One subscription, every feature, both platforms as they ship.
Your AI usage
$0 + / month, at cost
Paid straight to your provider, never marked up by us. Run everything on-device and this line is genuinely $0.
| Setup | Transcription | Summaries | Per hour |
|---|---|---|---|
| Fully offline | Vosk, on-device | Ollama / LM Studio, local | $0.00 |
| Hybrid — the sweet spot | Vosk, on-device | Claude Haiku 4.5 | ~$0.02 |
| Cheapest all-cloud | gpt-4o-mini-transcribe | GPT-5.4 nano | ~$0.18 |
| Accuracy-first | whisper-1 | Claude Sonnet 5 | ~$0.41 |
How these are worked out: one hour of conversational speech is roughly
9,000 words, or about 12,000 tokens of transcript, plus around 600 tokens of generated
title, summary, and extracted events. Transcription is billed per audio minute
(gpt-4o-mini-transcribe $0.003/min, whisper-1 $0.006/min);
summaries are billed per token at each provider's published rate. These are
estimates, not quotes — your real numbers depend on how densely people
actually talk and on provider pricing at the time, which is exactly why the in-app price
table is editable.
| How you'd use it | Setup | AI usage | Total / month |
|---|---|---|---|
| Any amount, fully offline | Vosk + local model | $0.00 | $5.00 |
| ~3 h/day of talk, hybrid | Vosk + Claude Haiku 4.5 | ~$1.50 | ~$6.50 |
| ~1 h/day of meetings | Cloud transcription + GPT-5.4 nano | ~$5.50 | ~$10.50 |
| Passive all day, ~2 h of actual speech | Cloud transcription + GPT-5.4 nano | ~$11.00 | ~$16.00 |
| Heavy — ~3 h/day, all cloud | Cloud transcription + GPT-5.4 nano | ~$16.50 | ~$21.50 |
Notice the pattern: transcription is the expensive part, and it's the part you can move on-device for free. That's why the hybrid row is a rounding error — Vosk does the heavy lifting locally and a small hosted model handles the writing. Note too that passive mode bills for detected speech, not for hours of runtime: voice activity detection means eight hours in your pocket is nothing like eight hours of billed audio.
The app works. The Android build is in testers' hands and the iOS port is underway. What gets this to a lot more people, a lot sooner, is the right people alongside it.
Integrations, distribution, and bundles. If you build for people who live in meetings, field work, journalism, accessibility, research, or healthcare admin — there's an obvious fit here, and an app deliberately built to talk to whatever endpoint you point it at.
Productivity, privacy, self-hosting, local-LLM, Android power-user, or second-brain audiences — this demos well and it holds up to scrutiny, because the offline mode is real and the cost tracker is honest. Early builds, direct access to the developer, and affiliate terms available.
A shipping product with a clear $5/month subscription, no inference costs on our balance sheet — users bring their own keys — and a second platform in development. Happy to walk through the roadmap, the architecture, and the numbers.
Why the economics are unusual: because SID3KICK is on-device-first and bring-your-own-key, we don't pay for inference and we don't operate a storage backend. There's no per-user cloud cost quietly scaling underneath the subscription — which means the $5/month is margin, not a loss-leader waiting to be repriced.
Partnership, coverage, investment, or you just want on the early-access list — it all comes to the same inbox and it all gets a reply.
Prefer plain email? [email protected]