Yaresaz isn't a chatbot bolted on beside the work. It's a layer that sits on your project's real data — messages, meetings, tasks and inventory — and what comes out is an assignable task, a set of minutes, a contract draft, a report. Not just a paragraph of text.
Most work is born in a single sentence in a conversation — and lost in the same place. Yaresaz picks that sentence up, analyses it in the background, and puts an editable task in front of you, with a suggested parent and a warning if something similar already exists. Nothing is created automatically; the last click is yours.
You tap "Suggest task" on that message.
You don't wait around. When a suggestion is ready, a small banner tells you.
The percentage on the card is exactly what the product shows: the model's own confidence in that one suggestion — not a stated accuracy figure.
This is the exact path the product takes, in the same order and with the same statuses. The work starts after the meeting ends, not during it.
Record the meeting in Zoom, Teams or any other recorder, then upload the audio file into Yeksaz. Yeksaz does not record the meeting itself.
Once the meeting is over, the file is transcribed and the conversation is split into speaker-labelled segments.
You map the detected speakers onto the meeting's attendees so the summary is written with real names. This step is optional and can be skipped.
The meeting content is indexed so you can search it and cite it later from a chat with the assistant.
A full summary, key points, decisions taken, action items and suggested tasks — each section separate and counted.
Every suggested task opens the task form in one click, pre-filled with title, description, owner, auditor and deadline — all editable before you save.
"Ask the project" searches your conversations, tasks, meetings and inventory, and answers with a link back to the exact message or meeting. "Technical advice" lays construction methods side by side — waffle slab against joist-and-block, for instance — with a high/medium/low fit rating, pros, cons and a recommended option. Compliance threads sit in the same list under their own badge.
"Waffle slab or joist-and-block for an 8m span?"
This comparison does not replace sign-off by a structural engineer.
Yaresaz does not read prices off the market. It takes the quotes your own team gathered — whether sourced internally or uploaded by you — extracts them and puts them in one table: pros, cons, a quality note, the differences that matter, and a recommended option with its reasoning. You make the final selection.
There is no manual form for a new contract: you say what you need and Yaresaz produces a draft — title, scope, body, amount, dates, payment schedule and milestones. Everything is editable in a review dialog before it is saved, the editor autosaves as you type, and "rewrite with AI" reworks the text to your own instruction — with the previous text restorable.
The product says this itself, and so do we: an AI draft is not a substitute for legal advice.
"I want to draft a new contract for the north facade work."
Weekly and daily reports are generated automatically, server-side. There is no "generate report" button in the product — users read them. The weekly report has eight fixed sections; the daily one is shorter and arrives with a handful of suggested tasks.
The daily report wizard opens with a snapshot of today — tasks completed, tasks overdue, stock issued and received, open purchase requests. Then it asks that project's own template questions, and answers can simply be spoken; speech becomes text that stays fully editable before you confirm. Walk away mid-way and the draft is kept — you resume from the same step.
"What held things up today?"
Ready-mix concrete didn't arrive until 11; the sixth-floor pour ran two hours late.
Messages are analysed one at a time in the background, capped at three pending suggestions per conversation, with a short gap between requests.
Before a task is created, the closest existing match is surfaced — same title or merely similar — and you can open it or knowingly continue.
If the message continues existing work, the suggestion is created as a subtask under that parent — and you can change the parent.
Assistant answers come with source chips, and each chip links straight to the task, message, meeting or file it came from.
A document with no extractable text is read visually instead, so it doesn't disappear from project search.
In a hurry, you can skip mapping speakers to attendees and the minutes processing carries on regardless.
Drafts save as you type and the save state is visible; once a contract is finalised it is no longer editable.
Attachments and voice messages are off in assistant threads. Text, an answer, and its sources — that's it.
One short session — we'll walk these exact flows on real data from a sample project.
Bring your project onto Yeksaz. In one short session, see a clear picture of the platform and the owner's assistant, working on your own real project data.