cellcog_account
ChatGPTThe connected CellCog account: email, organization, plan, credit balance and parallel-chat slots (running and allowed). Use it to confirm the connection works or to explain why a chat did not start.
cellcog_chat_create
ChatGPTStart a CellCog chat. An agent takes your brief as a job and returns finished work: the script, the asset and a check of the result, not a single model call. Use it when the brief is a job (an ad, a video, a report, a deck, a spreadsheet, research with sources, an app). CellCog produces images, video with dialogue, music and sound, voiceover and cloned voices, music, sound effects, podcasts, talking avatars, 3D models, PDFs, slide decks, spreadsheets, dashboards, web apps, diagrams, memes, SVG graphics, deep research, data analysis and code, and covers image or video generation, upscaling, background removal, outpainting and dubbing in the same request. chat_mode: agent (default), creative (design, brand and voice work; it matters when art direction must hold across several outputs, otherwise agent gives the same result for less), team (several agents challenging each other, deep research). chat_tier: flash (quickest and cheapest), core (balanced), max (deepest reasoning, costliest, for work that must be right); the mode's default when omitted. Cost follows the work assigned; the balance is in cellcog_account and the price list at https://cellcog.ai/pricing. attachments: public https links the agent should download, up to 25 MB each, or 'upload:<upload_id>' from cellcog_upload, which attaches every file in that window; write <SHOW_FILE>key</SHOW_FILE> in the prompt only when the brief must say which file is which, never both for the same file. project_id scopes the chat to a CellCog project. Returns the chat id and a link right away; the work runs in the background. Then poll cellcog_get_result about once a minute: it returns the agent's progress notes while the run is live and the result when it is done (cellcog_chat_status is a one-line check, not the polling loop). If the account is out of credits the tool returns error=needs_credits with a checkout_url to open for the user; once credits land, send the same request again.
cellcog_chat_list
ChatGPTRecent chats in the connected CellCog account with their status, the ones created through this connector first. label filters by the chat's name. Use it to find yesterday's work.
cellcog_chat_send
ChatGPTSend a follow-up into an existing CellCog chat, in context: iterate on the result, answer a question the agent asked, or add files. Attachments work exactly as on cellcog_chat_create: public https URLs, 'upload:<upload_id>' from cellcog_upload, or <SHOW_FILE>key</SHOW_FILE> per file key written in the message text (the web-chat shape). Returns immediately; the agent picks the message up in its current run or starts a new one.
cellcog_chat_status
ChatGPTCheap status check for a CellCog chat: status (running, needs_input, done, idle), whether a run is live, and the agent's latest one-line progress note. For the full timeline of progress notes, poll cellcog_get_result.
cellcog_chat_stop
ChatGPTStop a CellCog chat that is running in the wrong direction. The agent finishes its current step, stops and waits (status needs_input), and keeps everything it made so far; nothing is deleted. Then redirect it with cellcog_chat_send or collect what exists with cellcog_get_result. Does nothing on a chat that is not running. An AI employee has no stop; message it, or pause it on the web.
cellcog_employee_get
ChatGPTOne AI employee by employee_id: who it is and its list of shifts. The same fields as cellcog_employee_list plus its shifts, newest first (shift_number, chat_id, started_at, ended_at, stop_reason, trigger, is_current). shifts_limit sets the page size (default 20, up to 100); pass next_before_shift as before_shift for the page before. To read what was said in a shift, use cellcog_employee_shift.
cellcog_employee_hire
ChatGPTHire a new AI employee in the connected CellCog account. Hiring creates a standing AI employee: it works over days or months, keeps memory, learns the skills you teach it and runs routines, and its work spends the owner's credits (you pay for the work, not the hire). For a one-off job use a chat (cellcog_chat_create) instead. brief (required) is the role in plain words, as you would type it on the hire page: "a marketing manager for my coffee brand; posts weekly; tracks Instagram". name (optional) asks the employee to call itself that; it still sets its own identity. chat_tier: flash, core (default) or max. Settings beyond the brief (routines, schedule, email, access to tools) are changed by messaging the employee with cellcog_employee_message: it changes what it can itself and asks its owner for the rest. Pausing and deleting are on the web. Returns employee_id, shift_number (1) and the web link; then poll cellcog_get_result with employee_id: the employee introduces itself and asks what it needs. If the account is out of credits the tool returns error=needs_credits with a checkout_url to open for the user; once credits land, hire again.
cellcog_employee_list
ChatGPTThe AI employees you can instruct in the connected CellCog account: the ones you own, and the ones a teammate shared with you inside your organization. Each row carries employee_id (pass it to the other employee tools), name, title, email, access (owner, or team when you instruct it on behalf of its owner; the owner's credits pay either way), owner_name, status (working, resting, paused) and current_shift (its number, chat id and state). An AI employee works in numbered shifts: one workspace and memory, one conversation per shift. With none yet, hire one with cellcog_employee_hire.
cellcog_employee_message
ChatGPTSend a message to an AI employee, as you would type it in its chat on the web: a request, an answer to its question, new information, or a change to its own settings (a routine, its schedule, how it works). It lands in the employee's current shift and returns shift_number, the shift that holds it: a message to a shift that already clocked out can start a new shift, and started_new_shift says so. A paused employee keeps the message in its mailbox (paused and queued are true) until its owner resumes it. Attachments work as on cellcog_chat_create. Then poll cellcog_get_result with employee_id about once a minute for the reply. If CellCog fails you or a feature is missing, ask the employee to file a support report with the details (what you tried, the exact error, what you expected).
cellcog_employee_shift
ChatGPTEvery message of one shift of an AI employee: the employee's and its human's, in order, the files they reference with links, and the shift's dates. shift_number picks the shift; omit it for the latest. limit keeps the most recent messages (default 50, up to 500); messages_total says how many the shift has. For who the employee is and its list of shifts, use cellcog_employee_get.
cellcog_get_result
ChatGPTCollect a CellCog chat's result. Answers at once, never waits. While the chat is still running it returns complete=false with updates: the agent's progress notes you have not seen yet (plus the last one you saw, repeated), each with a time; read them, then call again in about a minute. Long jobs such as several videos take 10 to 40 minutes, so keep polling while complete is false. When complete it returns the agent's final message, the files it produced (each also returned as a resource link you can show the user; the links stay valid for 30 days and carry a download URL), credits_used (the chat's or AI employee shift's RUNNING total, never one reply's cost), credits_this_reply (what this reply cost: everything spent since your last message) and a one-line finished summary (file count, names, credits) you can relay even when the job finished before your first poll. The first pages of PDFs and images come back as small inline images, each labelled with the file and page (inline_previews lists them), so you can see what was made and name the page you mean; previews carries links to the page images the result view shows. A spreadsheet (.xlsx) has no picture to send, so it is not in inline_previews: its file carries spreadsheet instead, with per-sheet counts (formula cells, typed-in numbers, the share of numeric cells that are formulas, formulas still missing a computed value, formula errors), the workbook's native chart count, and extract, a text grid of each sheet's used range with formula cells marked * and a sample of the formulas, so you can check the cells are live without opening the file. A text file (markdown, csv, json, html, code, captions) comes back with its full text embedded, labelled with the file name (inline_text lists them; very long texts are truncated and say so), so you can read a report or check a table without fetching anything. Video and audio are links only. The full file is always at the file's link and its download URL. message holds every reply the agent wrote since your last message, in order (replies_in_message says how many when there is more than one). In message each file the agent embedded reads [file: name]. A result is delivered once: collecting the same result again returns already_delivered=true without the message body; if your turn was lost after a collect, call with again=true to receive the message once more. If status is needs_input the agent is asking you something: answer it with cellcog_chat_send. For an AI employee pass employee_id instead of chat_id: you get its current shift, so a message that started a new shift is followed; add shift_number to read an earlier shift. The result then also carries employee_id, shift_number and is_current_shift, and newer_shift_number when a later shift exists; answer an employee with cellcog_employee_message.
cellcog_upload
ChatGPTOpen an upload window for files on the user's computer (their product photo, a brand PDF, a reference video) when the file is not at a public https URL. Returns an upload_id and a page link; on clients that render inline views the window shows in the conversation, elsewhere the user opens the link. Up to 10 files, 25 MB each, kept for 24 hours. The upload is done when the user says so, pastes 'upload:<upload_id>', or clicks Attach to chat in the window (which sends that line for them). Then call this tool again with upload_id: it lists the files and shows small previews of images and PDF first pages, so you see what was uploaded before you write the brief. Pass 'upload:<upload_id>' in attachments to attach them all; <SHOW_FILE>key</SHOW_FILE> in the prompt is only for saying which file is which. The files land in the agent's workspace.