The 2025 report · public

The year we
got loud by getting quiet.

Krytz spent 2025 in private beta. We shipped 47 builds, talked to 180 users, and answered exactly one question over and over: what should I do right now?

47,238
Fragments captured / day
180
Beta users
99.97%
Uptime · 90-day
$0
Marketing spend
01 · A letter to people who use lists

We were quiet on purpose. Here's what we did while you weren't watching.

If you've heard of Krytz, you've probably also heard us complain — quietly, on purpose — that productivity tools are passive. They wait for you to organize. They wait for you to prioritize. They wait for you to be the system.

We don't think that's the right design. We think the tool should hold the picture. It should read the messy fragments, reconstruct the live state, and commit to one primary action — with reasoning attached, so you can trust it or tell it it's wrong.

In 2025 we tested that bet against 180 real users for 11 months. We rewrote the memory layer once. We rewrote the priority layer twice. We built an activity history that now backs every decision the system makes. The product is not done — it is, structurally, just becoming what we set out to build.

This page is an account of that year. Not a deck. Not a pitch. The actual numbers, the actual things we shipped, the actual moments that mattered. If we did this right, you should be able to feel the shape of the company through these pages.

— Maya & Lex, co-founders
02 · The numbers

The numbers that actually matter for a system like this.

Captures processed
14.2M
Every voice memo, paste, file drop, and inbox forward. ~78k captures per active user per year. The system parsed all of them.
Decision audit events
8.4M
Every rank change, primary commit, and "why-it-moved" trace. Stored permanently. Every one points back at the captures that drove it.
Capture p99
38ms
Local parser, on-device. Sub-50ms across iOS, Android, web. Never blocked on a network round-trip.
User-reported "feels confused"
0
After the anti-jitter work shipped in March. Down from 4 reports/week prior. We chase this metric quietly.
Systemers
4
Including the founders. We are not raising. We are not hiring. We are shipping.
03 · What we shipped

Forty-seven builds. Six surfaces. One closed loop.

Jan · Q1
Capture Console v1 + local NLU
The first time the system parsed a fragment in under 50ms without a network call. Tasks, dates, blockers extracted on-device. The shape of the product locked in.
captureon-devicemilestone
Mar · Q1
Anti-jitter primitives
Settling rules + cascade-depth cap + jitter guard. Cut rank churn 94% in our beta corpus. The day the system started feeling stable.
prioritizationstability
May · Q2
Decision Trace · the activity history
Every primary commit now points at the captures that drove it. Tap any task, see the 7-signal trace. The most-requested feature became the most-loved.
trustaudit
Jul · Q3
Memory rewrite · vector → chronological
Rebuilt the memory layer in 11 days. Faster recall, more accurate, and finally provenance-aware. The single biggest unblock of the year.
memoryrewrite
Sep · Q3
Pressure score · Strategy view
A single 0–100 number that summarizes your load with the lever to drop it. The metric most users said they "didn't know they needed."
strategy
Nov · Q4
Override · remember pattern
Correct the system once. The model remembers the rule for items of that kind. The day Asha (you'll meet her in the Voices section) stopped telling us we were dumb.
learning
Dec · Q4
Control Center · the governance layer
Inference limits, AI budget caps, export-anything / delete-anything. The boring layer that took six months because trust is boring.
governanceprivacy
04 · The shape of growth

We grew slowly on purpose. One user at a time, on a 30-minute call.

Beta cohort · cumulative · 12 months
● users · ○ daily-active
200 150 100 50 JAN MAR MAY JUL SEP NOV DEC ANTI-JITTER · MAR MEMORY REWRITE · JUL
05 · The voices

The people who put up with us for an entire year.

I have tried every productivity tool for the last decade. Krytz is the first one that doesn't ask me what I want to do — it tells me. And it explains why. After three weeks I stopped opening Things.
Mira Chen · founder · LIGHTPATH
Your priority layer is dumb. Fix it.
Asha Pradhan · day 22 · email at 11:48pm · (we did)
The fact that I can replay any day and see what the system thought at 3pm changed how I argue with myself about priorities.
Noor Bhatti · chief of staff · HEATHLAND
06 · 2026

The year we get loud on purpose.

Public beta opens Q1. Pricing locks in Q2. The first 10,000 users get the same 30-minute call we gave the first 180. Then we'll see if the thing scales.

Krytz · 2025 report · published May 14, 2026