Pricing, without the fog.
The desktop product is free for personal use, always. Paid tiers buy exactly one thing: our hosted inference — for people who'd rather not manage keys or models. Your plan's price is your runway; it can never surprise you.
Free
- The full local product — forever
- 100% local, or bring your own keys
- ~$0.10/day hosted taste (resets daily)
- Source code included (USL-1.0)
Budget
- One $2/mo hosted pool — no daily cap
- Economy routing: capable models picked for value
- Hard cap — never a surprise bill
Standard
- One $5/mo hosted pool
- Named, reliable models
- Web research on — live sources in the clipboard workflow
- Overage: pause, bump, or meter
Pro
- One $10/mo hosted pool
- Premium frontier models — and your pick
- Priority routing at peak times
- Overage flows into pay-as-you-go
What actually differs between the paid tiers
A fair question: if a plan just buys a pool of hosted inference, why not take Budget plus overage instead of a bigger plan? Because the tiers differ in three real ways beyond pool size:
- Routing. A "model" is the AI engine behind each request. To be direct: Budget is not a "worse summaries" tier — everyday quality is the same. Economy routing picks engines that do everyday work well for the least cost. Standard names specific, dependable engines. Pro unlocks the frontier engines for the hardest jobs — and lets you pick them yourself.
- Web research. Live research follow-ups call real search providers — a genuine per-request cost — so it's included on Standard and up. Every tier can do everything else; research simply makes the same features more powerful.
- Overage behavior. When your pool runs dry: Budget pauses or bumps to the next tier. Standard adds pay-as-you-go metering as a third choice. Pro is the top — there's nothing to bump to, so overage flows into metering automatically and you just keep working.
We're also evaluating faster premium endpoints — same answers, quicker — as a possible future Pro perk. Nothing to buy yet; we'll only ship it if it's actually worth it.
Wait — doesn't Free beat Budget?
On pure volume, sometimes yes! Free meters ~$0.10/day (≈ $3/mo of daily tastes) and it resets every morning — we keep it generous on purpose. Paid tiers buy three things Free doesn't: no daily cap (one monthly pool, spend it whenever), the tier features above (routing, web research, overage control), and the warm feeling of funding the work. If you'd honestly use less than ~$2 a month, stay Free — happily.
What each tier's runway gets you (measured, approximate)
| Action | $2 Budget | $5 Standard | $10 Pro |
|---|---|---|---|
| 📝 Summarize an article | ≈ 14,000 | ≈ 35,000 | ≈ 70,000 |
| 🎙 Professionally rewrite a dictation | ≈ 30,000 | ≈ 75,000 | ≈ 150,000 |
| ⏱ Transcribe speech with hosted inference | ≈ 40 hours | ≈ 100 hours | ≈ 200 hours |
| ⚡ Run an agent-assisted session* | ≈ 400 | ≈ 1,000 | ≈ 2,000 |
Free tier: ~700 summaries or ~2 hours of transcription per day, on the house.
About transcription: local vs hosted
Local Whisper in the app is free, unlimited, and private — and on a decent machine it is not slow. Hosted transcription isn't really about speed; it's about headroom: hosted runs a much larger Whisper model than most laptops can comfortably hold, so hard audio — heavy accents, background noise, long recordings — comes out cleaner, and there's nothing to install or keep updated. Roughly, an hour of hosted audio costs about as much as 400 article summaries.
How usage & overage work
Your plan's price is your runway. When it runs low you choose, per account:
Pause — usage stops at your runway until next month. A hard stop, never a surprise bill. · Bump — continue into one more block of your plan's size; we bill that block the moment it starts, and you're never floating more than one payment. · Pay-as-you-go — meter past your pool at the posted rates, for as long as you enable it.
On Pro, bumping has nowhere higher to go, so it flows into pay-as-you-go automatically — the top tier never interrupts you mid-work.
Bring your own keys and none of this applies — hosted blocks are optional convenience.
What a dollar buys (measured on our gateway, not marketing math)
| Action | per $1 |
|---|---|
| 📝 Summarize an article | ≈ 7,000 |
| 🎙 Professionally rewrite a dictation | ≈ 15,000 |
| ⏱ Transcribe speech with hosted inference | ≈ 20 hours |
| ⚡ Run an agent-assisted session* | ≈ 200+ |
Your $10 buys $10 of inference at our posted rates — the rates include a built-in 20% margin over our measured wholesale cost. We re-route requests between engines as prices and quality shift — always to keep that value honest — and your posted rate stays the same.
*Heavier agent sessions on premium models cost more; a typical session measures in fractions of a cent.