Who built it
By team researchers, for team researchers.
Marc Becker
Maastricht University
Stefan Berger
University of Groningen
Roman Briker
Maastricht University
Manuel Vaulont
Northeastern University
You script the team's conversation in advance. Three AI teammates play it back on schedule while one real participant joins in as a fourth member. Whenever the participant types, a backstage model — the director — rewrites the bots' upcoming messages so the team responds to what was actually said. Same manipulation for every participant, a real conversation for each.
conversationID captured
For researchers in management, psychology and organizational behavior who need team interactions that are comparable across participants: you define what the team says and when, the studio keeps it identical across sessions — and handles the one thing you cannot script, your participant.
Every message has a sender, a timestamp and a typing window — you design the exact rhythm: quick replies, long silences, slow typers. Every participant in a condition sees the identical conversation, so timing never becomes a confound.
When the participant sends a message, the director rewrites the bots' not-yet-sent messages so they answer it directly. It follows your written brief — task, personalities, condition — and never appears in the chat itself. Participants feel heard; your protocol stays intact.
The chat has four members. Blue, Green and Red are the bots that play your script; the participant always joins as Yellow. Everyone appears as a color only — no names, photos, or gender cues that could confound how the team is perceived.
The script is a list of messages: sender, text, timing. Edit it row by row in the visual editor or as JSON. Next to the editor sits a chat assistant: type "make Green more skeptical" and it rewrites the affected messages in front of you, with an undo for every change.
Ready-check modal, "teammates are joining…" countdown, task instructions and a session timer are per-project toggles. Reproduce your study flow exactly — say, 8 seconds of waiting, 10 minutes of chat, then back to the survey.
Every message is stored with millisecond timestamps, sender role and conversation ID. Export CSV for R or SPSS, rendered PDF transcripts for the appendix, or one ZIP with everything — TEST sessions stay tagged, so pilot runs never mix with real data.
A fixed script feels robotic the moment a participant types — so a second model, the director, manages the team backstage. It never writes to the participant itself. Its only job: rewrite what the bots are about to say, following the brief you give it — the task, each teammate's personality, your experimental condition. One full cycle:
Blue, Green and Red send their messages exactly when you scheduled them — typing indicators first, then the message.
Yellow contributes an idea, an objection, a question — something no fixed script could answer.
Chat so far, remaining script and your brief go to the model — back comes the rest of the conversation, rewritten to answer Yellow. Personalities and pacing stay as defined.
The rewritten messages replace the old ones mid-conversation. No pause, nothing to notice: the next teammate simply types a reply that fits.
To the participant it is a single natural team chat. To you, a session that followed the protocol — comparable across your whole sample.
Everything runs in the browser — for you and your participants, nothing to install. New accounts include a small demo project: open it, run a test chat, then reshape it into your study.
A project is one study cell: its own script, director brief, model and links. ID and password are generated for you; an optional account keeps them across devices.
Lay out who says what, when, with how much typing. Start from the demo script and adapt it — or let the AI assistant draft one from your task description.
Write the director's instructions: the task, each teammate's personality, your condition (e.g. high vs. low psychological safety). Pick provider and model; keys stay on the server.
Put the participant link in Qualtrics with a piped conversationID, pilot in TEST mode, watch sessions arrive on the review page — and export when you're done.
A chat assistant sits next to the script editor — and next to the director instructions. You type an instruction, it rewrites the affected messages directly in the editor, then lists what it changed. Every change has an undo.
Maastricht University
University of Groningen
Maastricht University
Northeastern University
Please cite the tool in your publications:
Create a project, script a conversation, run a test chat — your first working session takes minutes, not days.
Open the studio