The interview room

Watch them build the thing.

Not describe it on a whiteboard. Not type into a pad that cannot run. The call, the repository, a real terminal and a view of the candidate’s own AI, in one room — in place before anybody arrives.

Get early access Next: integrations

macOS app, Windows next · free for ten interviews a month, no card

The interview room for an architecture round: the repository's files down the left — docs, src, test, package.json, README — the round's brief open in the editor, the candidate on camera at top right with the questions panel beneath it, and the interviewer's own assistant along the bottom with example requests. — An architecture round, staged. The repository is on the left, the brief for the round is open in the editor, the candidate is on camera top right with the questions panel under her, and the interviewer's own assistant runs along the bottom. None of it was dragged into place.

What you have afterwards

An interview on a video call leaves one person’s recollection. This room records what happened as it happens, so an interview is something you can go back and read — and something a hiring manager can compare.

One transcript, one clock: Speech, agent activity and repository events are written to a single timeline, each attributed to the person who did it.
Test runs: Pass and fail per test, with the output kept underneath, so “it worked” is something you can open.
Scorecards: The write-up each interviewer files. Criteria live on the round, so two people who sat the same stage are comparable, and nobody sees another card before filing their own.

A shared workspace, not a shared screen

Screen sharing turns one person into a spectator watching a compressed video of somebody else's font. The room mirrors the work itself — the files, the editor, the terminal — so both people get a real, readable view of it on their own machine. Screen share is still there, as the backup for the things that are not code.

You see the work, not a picture of it: The candidate's editor and terminal are mirrored to the interviewer's machine as they type, readable at whatever size the interviewer likes.
The interviewer can help, with permission: Typing into the candidate's terminal — to pair on a stuck build — asks the candidate first. Until they say yes, nothing goes through.
The candidate brings their own AI: They sign in to an agent already on their machine. With Claude Code the interviewer sees what they asked it and what it did, as it happens. You can turn it off for a round.
Not only for engineers: A sales demo, a discovery call, a case study, a presentation, a data exercise, a support scenario — each opens with what that round needs, live or as a take-home.

Integrity, without a spyware pitch

Remote interviews have a real problem and most answers to it are worse than the problem. This one asks the candidate's permission, in plain words, and tells them exactly what it will and will not look at.

If an interviewer turns the checks on, the candidate sees the full list before anything runs and can decline — and declining is recorded rather than punished silently. It never reads their screen, their keystrokes, their files or any other application’s contents, and it cannot stop anything from running. Findings reach the interviewer as context, not as a score.

What the candidate sees →

Nothing to set up on the day

Every round points at a preset — a saved setup for one kind of interview — and the preset is what puts the room in place: the repository cloned on both machines, the panes the round calls for, the terminals opened in it, the brief pinned where the candidate can read it. Change it once and every interview at that stage changes with it, so two candidates who sat “the technical screen” sat the same thing, and the first ten minutes are the interview rather than setup. The interviewer walks in with questions already drafted by their assistant from the round’s criteria, the ones to hold back held back. Every change to a preset is kept as a version — history.

The presets screen, listing staged interview rounds with their repositories, durations and layouts. — The reusable unit. “Technical screen — bug fix” names a repository, a layout, the terminals to open and the brief to pin.

What a preset holds

The format: Ten are ready — video, coding, discovery, sales demo, presentation, case study, data analysis, support, customer success, custom. System design and whiteboard are shown in the picker greyed out, with the reason: they need the shared canvas, which is not built.
The repository, the terminals, the layout: Cloned on both machines before anyone arrives; the terminals the round calls for already open; the panes arranged for the round.
The brief: What the candidate reads, pinned in the room. The interviewer's assistant reads it too.
The materials, and who may see them: A scenario, a document, a dataset, a deck, a customer profile — each marked for the candidate, for the interviewer only, or held back to reveal during the interview. What the candidate is not meant to have never reaches their machine.
The criteria: What the panel rates, so a scorecard for this round asks the same questions of every candidate.
The candidate's AI, and the checks: Whether the candidate may use an agent for this round, and which integrity checks are asked for — the candidate sees both before anything starts.
Who runs it, and whether it may be practised on: The teams it belongs to, and whether a candidate may open it as a mock with a friend.

One preset opened: the room it stages — repositories, layout, terminals, preview, whose agent — its materials with their visibility, its criteria, the teams that run it, and its history. — One preset, opened. The room it stages, its materials and who may see each, its criteria, and — at the bottom — every version of it.

The interviewer's questions panel in the room: eight questions drafted from the round's rubric, the last three under a heading reading Hold these back, each with a button to reveal it to the candidate. — The interviewer’s side of the room. Questions drafted from the rubric before the call; the held-back ones go to the candidate one at a time, and the record keeps which ones were put.

For your engineers

How the room works

The mechanics, for whoever on your side is going to ask. What is mirrored and how, what needs the candidate's consent, and whose agent runs where.

Their editor, as they type: Content and cursor are streamed continuously, not on save. Follow mode moves the interviewer's view to whatever file the candidate opens next; it is a toggle, because reading ahead while they think is a normal thing to want.
Their terminal, live: The candidate's shells are mirrored to the interviewer's machine as real terminals, each keeping a tail of scrollback so a late join has the context rather than a blank panel.
Typing in their terminal needs a yes: Pairing on a stuck build is a real thing to want. The room asks the candidate first, and until they accept, nothing the interviewer types goes anywhere.
Screen share, as the backup: For everything that is not code — devtools, a simulator, a design file, a stubborn local setup. It gets a full pane rather than a thumbnail.

Whose agent

Two assistants can be in a room, and they are kept apart. The interviewer’s runs on an agent already on their machine — Claude Code, Codex CLI, Gemini CLI, Aider, opencode or Amp, whichever they use — and works only in the room’s own folder: the brief, the questions, the rubric, their notes, the deck or the sheet. It cannot reach the pipeline from there; nothing about your candidates is relayed to us. The candidate’s is theirs, and is one of these two:

The candidate's own: They sign in to an agent already on their machine — their subscription, their account. Nothing is billed to you and nothing is relayed through us. With Claude Code, every prompt, tool call and result is shown to the interviewer as it lands; the other agents run in the mirrored terminal, so their work is seen the way the rest of the work is.
Or none: Turn it off for the round. Some questions are better without one, and that should be your call rather than a limitation of the tool. When it is off, the candidate sees no assistant at all.

The room is the easy part to judge.

Stage one against a repository you know well. Ten a month, free, on the whole product.

Get early access Next: integrations