atypica_desk_research
ChatGPTDesk research: search the web and X (Twitter) for what is already known about a question and return a written summary in one call, which takes up to a few minutes. Present it as what was found online. No simulated consumers, no real people, no report page; for those start a study (atypica_study_create) or an interview (atypica_interview_create_project).
atypica_get_upload_credentials
ChatGPTFiles: get a presigned URL to upload an image or document. After uploading with HTTP PUT, pass the returned objectUrl where a tool takes a file: attachments of atypica_study_create or atypica_study_send_text_message, or images of an interview question.
atypica_interview_create_project
ChatGPTHuman interview: create an interview project in which real people answer structured questions asked by an AI interviewer. Each question can be open-ended, single-choice, or multiple-choice. Returns a projectId to use with other interview tools.
atypica_interview_generate_share_link
ChatGPTHuman interview: generate the link to send to people; each opens it to be interviewed by the AI interviewer. Supports permanent links or temporary links with configurable expiry.
atypica_interview_generate_stats
ChatGPTHuman interview: generate the aggregated statistics from completed sessions. This is an async operation (30-90s). Poll atypica_interview_get_stats to retrieve results.
atypica_interview_get_progress
ChatGPTHuman interview: how many people have answered: total sessions, completed vs in-progress, human vs AI persona breakdown.
atypica_interview_get_session
ChatGPTHuman interview: get one session in full: its structured answers, summary and transcript. A human interviewee is not identified (no name, account or personal info); an AI persona is named.
atypica_interview_get_stats
ChatGPTHuman interview: read the aggregated statistics per question across completed sessions. If stats are not yet generated, call atypica_interview_generate_stats first.
atypica_interview_list_projects
ChatGPTHuman interview: list the user's interview projects with brief, question count and whether stats exist.
atypica_interview_list_sessions
ChatGPTHuman interview: list one project's sessions (one per person) with pagination. Returns lightweight session info (title, status, type). Use atypica_interview_get_session for full details including structured answers.
atypica_interview_update_questions
ChatGPTHuman interview: replace the question list of an interview project. Provide the full updated list — existing question IDs are preserved when matched. Questions must follow validation rules: choice types need ≥2 options, open type must not have options.
atypica_panel_search
ChatGPTUser research: search the user's persona panels by title. A panel is a curated group of AI-simulated consumers to run a study with. Without a query, returns all panels. Use a panel ID with atypica_study_create to start a panel-based study.
atypica_persona_get
ChatGPTUser research: get one AI-simulated consumer (persona) in full, including its prompt.
atypica_persona_search
ChatGPTUser research: search AI-simulated consumers (personas) by text, optionally only the caller's private ones. Without a query, returns the latest available personas visible to the caller.
atypica_study_create
ChatGPTUser research: start a study in which AI-simulated consumers are interviewed and discuss, ending in a report. Pass the user's question as the first message. Returns the study's userChatToken, which every other study tool takes. The research agent starts at once and runs in the background, usually half an hour to an hour once the plan is confirmed; poll atypica_study_get_messages and follow its next field.
atypica_study_get_messages
ChatGPTUser research: read a study's conversation and what to do next. next.kind is one of: running (poll again in a minute or two), tool_result (the agent asked the user something: show next.questions or next.plan to the user, then atypica_study_submit_tool_result with next.toolCallId), artifact_ready (a report or podcast is finished: atypica_study_get_report or atypica_study_get_podcast with next.token), text_message (the agent spoke and stopped: relay it, reply with atypica_study_send_text_message), stalled (nothing running and nothing to answer: send a text message to start the agent again). steps lists what the agent has done so far, one entry per step (kind: social_observation, personas, interviews, discussion, audience_call, desk_research, report, podcast) with its state, what it was asked (brief), what came out (summary), facts in details (the personas built, searches per platform) and, for interviews, a discussion or a social observation, records: each with its own state and a stepToken. Tell the user what is happening instead of only that it is running. atypica_study_get_step reads what was said in a record by its stepToken, live. replayUrl opens the study's replay without sign-in: the user can watch the interviews and discussion there, reloading for updates.
atypica_study_get_podcast
ChatGPTUser research: get a finished podcast (script, audio URL, share URL) by its podcast token: next.token from atypica_study_get_messages when next.kind is artifact_ready with artifactType podcast. Not the study's userChatToken.
atypica_study_get_report
ChatGPTUser research: get a finished report (title, HTML, share URL) by its report token: next.token from atypica_study_get_messages when next.kind is artifact_ready with artifactType report. Not the study's userChatToken.
atypica_study_get_step
ChatGPTUser research: read what happened inside one step of a study, by a stepToken from steps in atypica_study_get_messages: the posts a social observation found, what the interviewer and one persona said, or what each persona said in a discussion. Works while the step is still running; pass after to get only what is new. Use it when the user wants to follow or dig into a step.
atypica_study_list
ChatGPTUser research: list the user's studies, newest first. Each token is that study's userChatToken, to resume it with the other study tools.
atypica_study_send_text_message
ChatGPTUser research: write a text message from the user into a study and run the research agent: a follow-up, a correction, the user's reply to something the agent said, 'please continue'. Returns at once; the agent runs for minutes, poll atypica_study_get_messages. Rejected while the agent is running. A question the agent asked as a tool call (next.kind tool_result) is answered with atypica_study_submit_tool_result instead.
atypica_study_submit_tool_result
ChatGPTUser research: give the research agent the result of a tool call it left for the user, reported by atypica_study_get_messages as next.kind tool_result (tool-requestInteraction: the user's choice among the options; tool-makeStudyPlan: whether the user confirms the plan). The output is the user's decision: show the question to the user and submit what the user chose, never your own guess. The agent then continues.