Local-first · AI-native · One-time purchase
Notes that think with you.
On your machine, and nowhere else.
Smritsavant is a spatial notepad with a semantic memory: every note is embedded, auto-linked, and answerable in chat — by AI models that run entirely on your device. Your email starts the trial — nothing else. No subscription, ever.
Signed builds are being prepared. The button emails us — we reply exactly once, on launch day.
- Works offline
- No telemetry — enforced by an automated test suite
- Your notes never leave your device
Why Smritsavant
Built on four promises.
Each one is a mechanism you can verify, not a slogan.
Private by architecture
The AI that reads your notes runs on your device. Cloud models are opt-in, bring-your-own-key, and gated behind a consent screen that names the exact endpoint. A privacy test suite fails the build if any other network call sneaks in — and an in-app Network Audit shows every gated request.
How privacy is enforced →Your data is yours
Notes live in a plain SQLite file on your disk. Export any canvas to Markdown, HTML, JSON, or PDF; back up the whole library to a single archive; or script it with the export CLI. Leaving Smritsavant is a supported feature.
Import & export docs →Notes that think with you
Every note is semantically embedded and auto-linked as you write. See your thinking as a spatial canvas, a firing Cortex, or a Chronicle timeline — and ask your notes questions in chat, with citations. AI edits always land behind a diff you review.
See the memory system →One purchase, yours forever
No subscription. A license is a one-time purchase, and every dollar you've paid counts toward the $29 Cross-Platform tier — forever. Minor updates are free; a major version is a new purchase, so development gets funded without renting your notes back to you.
How licensing works →Capabilities
The whole toolkit.
Spatial canvas
Blocks on an infinite canvas — merged, grouped, and linked by meaning, not folders.
Semantic search
Full-text plus meaning-based search in 50+ languages — entirely offline.
AI chat, your engine
Ask a canvas questions, get cited answers — local models on your own GPU/CPU by default.
Want a frontier model? Opt in with your own cloud API key — chat only; voice, search, and captions stay on-device.Voice
Dictate, capture lectures, drop in audio — transcribed on-device, searchable like text.
Web clipper
Save any page from your browser — article, selection or link. Pictures come with it and are described on your device, so you can find them later.
The extension has no network access: your clip travels to the app over a local connection and never touches a server of ours.Sync, your way
Any cloud folder or WebDAV server you already trust. No Smritsavant server in the middle.
Mobile
iPhone, iPad, and Android companions with the same private core — not yet in the app stores.
On-device AI chat isn't available on Android — bring your own cloud key there.Local AI, sized to your machine
Every machine gets a tier. Zero configuration.
Smritsavant detects your hardware and matches the AI to it — from phones to workstations. Each tier is a carefully chosen open model, pinned per release and upgraded over time as better open models ship. No model zoo to browse — the most you'll ever decide is a single speed-versus-quality choice on the largest tiers — and if you want more horsepower than any local tier gives, bring your own cloud key anytime: a far larger frontier model, behind an explicit privacy-responsibility disclaimer.
| Tier | Hardware | What you get |
|---|---|---|
| Ultra-Lite | Phones & tablets | On-device intelligence built into your device — chat via Apple Intelligence on iPhone/iPad; search, voice, and captions on-device everywhere. |
| Floor + Voice | 8 GB Macs · PCs below the speed bar | No machine is left out: voice notes with on-device transcription (queued rather than live on the slowest PCs), image captions in idle moments, full semantic search. Chat runs through your own cloud key — or Apple Intelligence where supported. |
| Lite | Apple Silicon 16 GB · any 8 GB+ GPU · PCs that pass the speed check | A compact ~4B-parameter model for capable chat on modest hardware — answers stay grounded in your notes. |
| Standard | Apple Silicon 24 GB · 12 GB+ GPU | The everyday sweet spot: a ~9B-parameter model with fluent chat, fast inline AI, comfortable long sessions. |
| Moderate | Apple Silicon 32 GB · 16 GB+ GPU | On Apple Silicon, a real jump: a ≈26B mixture-of-experts model that fits like a big model but answers at small-model speed. On 16 GB GPUs, the Standard brain at higher numerical precision — cleaner long-form output, same fast feel. |
| Supreme | Apple Silicon 36 GB · 24 GB+ GPU | Heavyweight ≈27B-class local intelligence — long-context synthesis and consolidation shine here. On Apple Silicon it’s a quality opt-in; the faster Moderate stays the default. |
| Ultra | Apple Silicon 48 GB+ · 32 GB GPU | The top tier, and the first with a choice: a fast ≈35B mixture-of-experts by default, or a dense quality option (≈31B on Apple Silicon; ≈27B on GPUs). Workstation-class answers, zero cloud. |
Honest expectations: answer quality and speed scale with the tier your machine can run — that's the physics of local AI, and we'd rather show you a tokens-per-second badge than hide it. An on-device model is genuinely capable for notes-grounded work, but it is not a frontier cloud model; if you want frontier-class answers, connect your own cloud key and Smritsavant will use it under the same consent-and-audit rules. On PCs without a dedicated GPU, what limits local chat is memory speed, not memory size — so instead of guessing from a spec sheet, the app runs a ten-second speed check on first launch, and that measurement decides how far up the ladder the machine goes — fast unified-memory PCs climb the same rungs as the table above, typical laptops run Lite, and machines below the bar keep the full voice, caption, and search experience — with transcription queued instead of live, and chat through your own key — rather than pretending a big model would run well. Voice and language coverage never shrink with hardware; only chat does.
Transparency note: tiers run openly-licensed models, downloaded from their official sources to your device. The exact model behind each tier in the current release is always listed in Models & licenses.
Pricing
Pay once. Everything counts.
Perpetual licenses, sold plainly. Every purchase you make is remembered as credit toward the top tier.
Android
- Android phone + tablet on your Google account
- Search, voice & captions on-device; chat via your own cloud key
iOS
- iPhone + iPad on your Apple ID
- Full Pro — AI via Apple Intelligence on supported devices*
Single Desktop
- One Mac, Windows, or Linux machine
- Full Pro feature set, on-device AI included
Cross-Platform
- 2 desktops + 3 mobile devices, any mix of platforms
- Everything, everywhere you work
* On iPhone/iPad, AI chat and image understanding use Apple Intelligence — Apple's foundation models, under Apple's on-device and Private Cloud Compute privacy guarantees — and need an Apple Intelligence-capable device (A17 Pro-class iPhone or supported iPad). Older devices can use a cloud model with your own key. On every platform, cloud AI is opt-in and chat-only — search, voice, and captions always run on-device.
The loyalty math
Upgrading to Cross-Platform never costs you what you've already spent:
Every non-refunded purchase on your account counts — including App Store and Play purchases you link by email. No waiting periods, no clawbacks, no fine print. Arithmetic.
| Started with Single Desktop ($10) | $19 to upgrade |
| Android + iOS linked ($18.99) | $10.01 to upgrade |
| Two desktops ($20) | $9 to upgrade |
| Paid $29 or more over time | Free |
Is this a subscription?
No. Every license is a one-time purchase that keeps working forever, with free updates for that major version. When a future major version (say, 2.0) ships, it's a new purchase — that's what funds development without renting your notes back to you. The version you own keeps running indefinitely and receives fixes until 90 days after the next major release.
What happens when my trial ends?
After 14 days the app becomes read-only. Nothing is deleted, nothing is held hostage: you can still open, search, export, and delete everything, forever. Buying a license re-enables editing on the spot.
How many devices can I use?
Each license adds device slots to your account, and they stack:
| License | Desktop slots | Mobile slots |
|---|---|---|
| Android | 0 | 1 |
| iOS | 0 | 1 |
| Single Desktop | 1 | 0 |
| Cross-Platform | 2 | 3 |
You can see and deactivate devices anytime from your account page.
Can I get a refund?
Direct purchases: 30 days, no questions asked — email[email protected]. App Store and Google Play purchases follow the stores' own refund processes, which we can't override.
Do I need an account?
An email address, yes — it's how your trial, license, and devices are tied together (sign-in is by magic link; there are no passwords). Your account holds license records only. Your notes are never in it.
Trade-offs
What Smritsavant is not.
- Not a collaboration tool. Sync connects your devices. There's no shared workspace, no real-time co-editing, no comments from teammates. If your work lives in shared team docs, a cloud workspace is the right tool — and we're not one.
- Not a cloud service. There's no web app and no server copy of your notes — that's the point. It also means backups are your responsibility (the app makes one-click archives, and any cloud folder you already use can carry sync).
- Not light on hardware. Local AI means real model downloads (a few hundred MB at the floor, up to ~22 GB for the biggest tier) and real compute. Chat speed and quality scale with your machine; older hardware will feel it — though voice, captions, and search run on every machine. On Android there's no local chat at all — you bring your own cloud key.
- Not a plugin platform. No extension marketplace, no community themes. What's in the docs is what it does — built and supported by a two-person team that answers its own support email.
If any of these disqualify us, better you know in five seconds than after a purchase.The comparison page names the well-known alternatives and says plainly when one of them is the better pick.
Try it where it runs: on your machine.
Signed builds are in preparation · no waitlist database — just one launch email