create_folder
ChatGPTCreate an owned folder, optionally nested. Omit parent_id for the root. Only create when requested, including 'save in a new folder'. First list folders to reuse an existing match. After an uncertain response, list before retrying.
delete_folder
ChatGPTDelete only the empty folder the user requested. Never deletes documents or subfolders. Non-empty folders are rejected; do not empty them automatically to make deletion succeed.
fetch
ChatGPTRead a source document by ID, including a completed transcript. Returns up to 24,000 characters. If metadata.next_offset is set, use read_document_page. Only documents owned by the connected user can be read.
get_supported_sites
ChatGPTList the websites whose recording pages Sofer.Ai currently accepts for transcription. Use when asked which sites work, or when unsure whether a page is supported. Returns site names and URLs from the same service as GET /link/sites. Pass a supported site's original recording URL to transcribe_audio_link; do not extract or download the audio yourself. This tool does not create a job or use transcription balance. Direct HTTPS audio links also work.
get_transcription_preferences
ChatGPTRead the connected account's saved Hebrew word formatting preference. Transcription automatically applies it when the format is omitted; no lookup is needed before each job. This preference also applies to website uploads.
get_transcription_status
ChatGPTCheck an existing Sofer.Ai transcription job. Fetch its text after COMPLETED. If still pending or processing, return its Sofer.Ai link and check again when requested. Report FAILED, CANCELLED or INSUFFICIENT_FUNDS without resubmitting.
list_documents
ChatGPTList active owned documents with titles, IDs and folders, even if not in the knowledge base. Omit folder_id for all active documents; provide it for direct contents of one folder (not its subfolders). Follow next_cursor for more. Use fetch to read text.
list_folders
ChatGPTList owned folders and IDs. Omit parent_id to list all folders, including nested ones. Follow next_cursor to find matches before creating a folder. Reuse a matching existing folder. Ask which parent/path when names are ambiguous; never guess IDs.
move_document
ChatGPTMove an owned document into an owned folder, or send null for root. Resolve IDs from list_documents/search and list_folders. Does not retranscribe, change content or charge transcription fees. Filing does not add to the knowledge base.
move_folder
ChatGPTMove an owned folder and its contents beneath another owned folder. Explicit null parent_id moves it to root. Does not rename it. Resolve both folders first; moves that create a cycle are rejected.
read_document_page
ChatGPTContinue reading a long document using the next_offset from fetch or the previous page.
rename_folder
ChatGPTRename an owned folder without moving it. Resolve its ID using list_folders.
search
ChatGPTFind passages relevant to a question in the connected user's Sofer.Ai Maishiv knowledge base. Connector searches have no Sofer.Ai search fee and do not consume the website's daily question allowance. Returns passages, source URLs and available audio start/end times in seconds; null times mean no reliable audio alignment. Fetch useful sources and cite them when answering. Empty results mean no evidence was found; do not invent source content.
set_default_hebrew_word_format
ChatGPTSave a default only when the user explicitly asks to remember it for future recordings. Updates their Sofer.Ai account, including website uploads. hybrid transliterates short phrases and uses Hebrew letters for longer ones; he uses Hebrew letters; en uses English transliteration. Null resets to Hybrid. Applies within English recordings. A one-recording request must use the transcription override instead. Existing transcripts are unchanged. Report the saved preference to the user.
transcribe_audio_file
ChatGPTTranscribe a user-selected audio attachment using their Sofer.Ai balance. Pass the real ChatGPT file object, never an invented file ID or local path. Sofer.Ai copies the file before queueing; report the job ID, returned settings and check status. Omit hebrew_word_format to use the saved account preference. Overrides do not change that preference. For non-English recordings only he is supported. For speaker labeling, set num_speakers to a known count above 1, or auto_detect_speakers=true when unknown; never combine them. Omit both for one speaker without labels. Changing speaker settings creates a new paid job; it does not modify an existing transcript. Identical settings reuse the job for 24 hours. Reusing a job leaves its folder unchanged; use move_document to refile it.
transcribe_audio_link
ChatGPTStart a paid transcription from a supported-site page or direct HTTPS audio link. Pass the user's original recording page directly, including YUTorah, TorahAnytime, MyTAT, AllDaf and YouTube. Call get_supported_sites for the current list when unsure. Sofer.Ai's transcription endpoint extracts the audio from supported recording pages. Unsupported pages require a direct audio link or attachment. Submit one recording, not a playlist. Use only when the user requests transcription. Uses their existing Sofer.Ai balance. Return the job ID; check status later, and do not retry an uncertain submission. For 24 hours, the same site URL, language, effective format and speaker settings reuse a job; direct audio is deduplicated by content. Different URLs may create separate jobs. Omit hebrew_word_format to use the saved account preference. Overrides do not change that preference. For non-English recordings only he is supported. For speaker labeling, set num_speakers to a known count above 1, or auto_detect_speakers=true when unknown; never combine them. Omit both for one speaker without labels. Changing speaker settings creates a new paid job; it does not modify an existing transcript. Identical settings reuse the job for 24 hours. Reusing a job leaves its folder unchanged; use move_document to refile it.