Aiko Tanaka18:20 Osaka
SI-T-20 is in. One shared context, 3 voices. Kofi, can you call audio.tick() from the loop?SI-T-20 Audio foundation · Done · 60%
Your studio's board, design and publisher reports, kept true by AI that reads, watches and listens to the work. Nobody updates a ticket. Nobody writes a status.
Every card is moved by hand, so the board drifts a week or a month behind reality. Everyone quietly stops trusting it.
People wait for someone to write a ticket before they start, or work off the books where nobody can see it.
When a publisher asks how it's going, the founder rebuilds the answer from memory and Slack scrollback, in the busiest week of the month.
Commits and changelists, chat and email, but also meeting recordings, playtest footage, nightly screenshots and the art itself, with the whole project's history in view at once. Each one becomes a claim with its evidence attached.
audio.tick() from the loop?All three cues are finished and in the repo. The explosion can't be wired until there's a hit event to listen to, so SI-T-21 is blocked on SI-T-13 for me.
No key in the comment. KTT looked at the sprites and matched them to the sprite-set task.
Click any field and see why it says that: the commit, the message, the moment in the recording, which reader made the claim and how sure it was.
When a person corrects a field, they pin it. No agent can change a pinned field until a person releases it.
Lena takes the invader animation as soon as formation lands.
The design lives as Markdown in git. When a meeting or the code changes something, KTT drafts the edit and asks the owner of that area in one message. "yes" merges it, and any work built for the old version is flagged.
No assignee, no status, no queue. Small changes apply with an undo, the rest arrive as one digest at the start of each person's day.
Publisher milestone reports, backer updates and status decks are generated from the log when you need them. Every line cites its evidence, and every report is frozen once it's sent. Promises made in them are tracked until they're kept or missed.
| Promise | Made | Status |
|---|---|---|
| A playable build for the showcase on 25 Sep | Email to Pixel Harbor, 15 Sep | Kept |
| Full player death and respawn | Showcase report, 25 Sep | Open |
| Vertical slice by March | Backer update, Aug | At risk |
Nobody tracks what they promised a backer six updates ago. KTT does, from the moment it's sent. (Example studio and promises.)
One click from any field to the commit, message or moment in a recording behind it.
A person's correction is permanent until a person releases it. No agent quietly undoes it.
What the AI may do alone is set per action by measured accuracy. Every correction counts against it.
Every raw signal is kept. When a better model arrives it re-reads the whole history and everything rebuilds.
KTT never shows hours, activity or "last seen". Your studio decides what it may record, and nothing it records is used to measure people.
We played a three-week Space Invaders sprint through KTT: 39 commits, 18 chat messages, 5 emails and 9 meetings from a four-person team, one of them remote. Then we graded the board every evening against what really happened.
real changes reached the board
of them the same day
owners right on the final evening
A simulated studio with an invented team, so treat these as directional, not a benchmark. The same test showed what to fix next, and we're fixing it.
Each source has one adapter that stores what it reads exactly as it arrived, so swapping a tool never touches your history. On Atlassian or Microsoft tools? KTT connects to those too, without ever depending on them, and it isn't tied to any one AI vendor.
We're working with a small number of studios on early access. Tell us about your team and what your board looks like today.