Does the AI decide who is rejected?

No. The workflow places a checkpoint before any rejection, because an application dismissed by mistake does not come back. The agents state the gap to the role, criterion by criterion, with the passage they rely on; the decision stays with the recruiter.

How do we avoid discriminatory criteria?

By writing them down in a document the agents re-read on every request. The file /grids/exclusions.md lists what must never enter an assessment, and it is visible and correctable — unlike an opaque filter whose rule nobody can verify.

Do applications leave the company?

They stay in your drive; our database holds no copy, only technical indexes. The extracts needed for an assessment go to the model provider you chose, on your own key — you decide which, including a European provider.

Recruitment

Assess against your criteria, not against the impression a CV leaves

Three recruiters read the same pile and shortlist different people, because the scoring grid only exists in conversation. And the published job description describes a role the team has already moved on from.

Create my first project

You describe your work

Nothing to configure. The blueprint derives the agents, their speciality and the folders they need — and submits the lot before writing anything.

Ts-Agents — Blueprint for “Hiring — Data engineer”

Your work

"We hire technical profiles. I want every application confronted with the same grid, and the gap to the real role spelled out — not a global score."

Proposed organisation · nothing written to your drive yet

3 agents proposed

Application analyst/applications
Confronts the background with the job requirements
Grid keeper/grids
Applies your criteria, with their weighting
Interview preparer/roles
Proposes the questions that resolve remaining doubts

Apply to projectEdit

The workspace takes shape

What you upload and what has to be written together, kept apart. The folders carry the words of your trade — never /documents.

Ts-Agents — Workspace

To upload

The folders are created in your drive. You put your documents in them.

/applications/<role>34 files
CVs and cover letters received, per open role
/roles6 files
Job descriptions, as the team actually describes them
/interviewsempty
Notes from past interviews

To write together

These documents do not exist yet. The assistant asks you questions and writes them.

/grids/data-engineer.mdwritten
Assessment criteria and their weighting
/grids/exclusions.mdto write
What must never enter an assessment

Write this document

You ask, the team works

Every step is visible while it runs: the file read, the specialist consulted. The answer cites the paths it relies on.

Ts-Agents — Conversation

Which profiles in the pile meet the blocking criteria for the role?

The organisation is working — 4 steps

list_folderlists /applications/data-engineer
read_filereads /grids/data-engineer.md
ask_agentconsults the grid keeper
ask_agentconsults the application analyst

Seven profiles out of thirty-four pass the blocking criteria. Two deserve a nuance: their real-time experience does not appear in the CV but in the cover letter…

Where the AI stops and asks

These checkpoints are written into workflow.md, in plain text. Between two of them the AI works alone; it does not cross one. You decide where they fall.

Before any rejection: an application dismissed by mistake does not come back.
Before adding a criterion to the grid, which will apply to every candidate.
Before sending an interview question to a candidate.

The project remembers

Your decisions carry from one conversation to the next — and you can read them, correct them, erase them. A memory you cannot contradict is not one.

Ts-Agents — Project memory

What the project retained4 entries

Re-read on every request. Correct what is wrong, erase what no longer serves.

## Decisions

- Streaming experience is blocking; the degree is not.
- Never take age, name or address into account: see /grids/exclusions.md.
- A profile rejected in 2025 may reapply — do not disqualify automatically.

## Preferences

- State the gap to the role, never a global score.

SaveForget

Frequently asked questions

Does the AI decide who is rejected?: No. The workflow places a checkpoint before any rejection, because an application dismissed by mistake does not come back. The agents state the gap to the role, criterion by criterion, with the passage they rely on; the decision stays with the recruiter.
How do we avoid discriminatory criteria?: By writing them down in a document the agents re-read on every request. The file /grids/exclusions.md lists what must never enter an assessment, and it is visible and correctable — unlike an opaque filter whose rule nobody can verify.
Do applications leave the company?: They stay in your drive; our database holds no copy, only technical indexes. The extracts needed for an assessment go to the model provider you chose, on your own key — you decide which, including a European provider.

Faithful reconstructions of the interface, built with the product's real components and colours. These are not screen photographs.

Other trades

Describe your work. See what we make of it.

The blueprint shows you the proposed organisation before writing anything to your drive. Judge it on the evidence.

Create my first project