get_action_items
ChatGPTList the signed-in user's own action items (to-dos, read-only). Use this when the user asks "what are my to-dos" or "what is still pending". items is one merged list of two sources, treated alike: (1) structured to-dos already created in the account, and (2) action items extracted by rules from recording summaries (summary body; falls back to the mindmap when no action-items section exists). Each item carries source (memo = structured to-do; main_text/mindmap = extracted from a summary) and completable (true = a real to-do with an itemId that can be completed or reminded in the app; false = an action item merely written in a summary, no to-do created yet, cannot be completed or reminded - always say so when replying). Optional recording_id restricts to to-dos derived from one recording plus action items extracted from its summary; omit it to list everything (do not pass it when the user asks for "my to-dos"). By default only open items are returned; include_completed=true also returns completed ones. due_from/due_to filter remind_time and created_from/created_to filter create_time (all yyyy-MM-dd, inclusive, UTC+8); any date filter returns structured to-dos only (extracted items have no dates). With no to-dos, returns an empty items array and total 0 (a normal response, not an error). hasActionItems in list/detail is only an on/off flag; call this tool for the actual items.
get_map
ChatGPTGet the existing mindmap source from a recording's summary (read-only, never triggers generation). recording_id is required; use the recordingId value from list_recordings / search_recordings. Returns the mindmap source (markmap-compatible Markdown) that can be rendered directly, plus updatedAt. This tool only returns an existing mindmap and never generates/regenerates; to generate/regenerate, use the Boya OS app. If there is no mindmap (summary not generated or no mindmap content), returns hasMap=false and an empty mindmap string (a normal response, not an error). When you need the full summary (body/points/mindmap together), prefer get_summary - the mindmap is just one of its fields.
get_recording
ChatGPTGet one recording's details (read-only). recording_id (returned as recordingId) is required; use the recordingId value from list_recordings / search_recordings. Returns title, file name, file type/extension, folder, duration durationSec (seconds), language, transcript status, derived flags hasTranscript/hasSummary/hasMindmap/hasActionItems, recorded time recordedAt and created time createdAt. With include_audio_url=true it also returns audioUrl: a 10-minute pre-signed playback link that must be re-fetched after it expires. If the recording is missing or not accessible, returns isError RESOURCE_NOT_FOUND (anti-enumeration: no distinction).
get_summary
ChatGPTGet one recording's latest existing AI summary (read-only). recording_id is required; use the recordingId value from list_recordings / search_recordings. Returns title, summary body summaryText, key points mainText, keywords, mindmap source, version, and updatedAt. One call gives both summary and mindmap; use get_map if you only want the mindmap source. This tool only returns the latest existing summary and never triggers generation or regeneration; to generate/regenerate, use the Boya OS app. If there is no summary or generation failed, returns hasSummary=false with content fields null (a normal response, not an error), which you can use to tell the user the summary is not ready. For batch use, first filter recordings with hasSummary via list_recordings to avoid ineffective calls.
get_transcript
ChatGPTGet one recording's transcript (read-only). recording_id is required; use the recordingId value from list_recordings / search_recordings. format is one of text (default, readable plain text), markdown (structured, easier to process), json (raw per-sentence data with speakers/start-end times/text; most complete and not truncated). For text/markdown, add include_speakers / include_timestamps for speaker labels and [mm:ss] timestamps; for long transcripts paginate by sentence with offset/limit (limit 1-50, default 50), which returns total and hasMore; max_chars caps a single page and sets truncated=true when exceeded. To view a time range, use start_ms/end_ms (milliseconds, inclusive) as a window; only sentences overlapping the range are returned, and sentenceCount/content reflect the current page within the window. Returns sentenceCount (this page), total (total sentences in the window), offset, limit, hasMore, language, durationSec. If transcription has not succeeded, returns a non-success status with empty content (a normal response, not an error); to check whether a readable transcript exists, first look at hasTranscript from list_recordings.
list_devices
ChatGPTList the current account's bound recording devices (read-only). Use it to learn "which devices exist"; returns each device's deviceId (the device serial), deviceName, sn, firmwareVersion, and boundAt. No recording count is returned, and recordings cannot be filtered by device (the device ownership column is not populated on every upload path, so such a filter would report nothing). Only devices in the bound state are returned; unbound devices are excluded.
list_folders
ChatGPTList the current account's recording folders (read-only). Use it to learn "which folders recordings are in and how many each holds"; returns each folder's folderId (usable directly as folder_id in list_recordings), name, parentId, createdAt, and recordingCount (the account's non-deleted/non-recycled recordings directly in that folder). Recordings that are uncategorized (root) belong to no folder and are not listed; to see the recordings in a folder, pass folderId to list_recordings.
list_recordings
ChatGPTList the signed-in user's own recordings (read-only, newest first by recording time). With no filters it returns the latest page. Filter by date range (date_from/date_to, yyyy-MM-dd, inclusive, by recording time) or a relative period (period: today / yesterday / this_week / last_week / this_month / last_month; mutually exclusive with date_from/date_to), or folder (folder_id); use min_duration/max_duration (seconds) to keep only recordings whose duration can be determined. Paginate with page / page_size (page_size 1-100, default 20); total is the total match count, so keep paging while fewer than total rows are returned. Each item has recordingId (pass it as recording_id to get_recording / get_transcript / get_summary / get_map / get_action_items), title, folderName (null at root), durationSec (seconds), language, status (pending/processing/success/failed), derived flags hasTranscript/hasSummary/hasMindmap/hasActionItems (whether readable content exists), and time fields.
search_recordings
ChatGPTSearch the user's own recordings by keyword (read-only). This is the go-to tool for "I don't remember which recording, but I recall X was said in it". Keywords match literally (substring) against title/file name, transcript, and AI summary; this is not semantic/fuzzy search, so use words or phrases that may appear verbatim in the text (a single keyword). scope limits the sources: all=everything (default), title=title/file name, transcript=transcript, summary=summary; leave it default when unsure. Each recording keeps only its highest-priority match and reports the source matchType: title > transcript > summary, plus a snippet. Paginate with page / page_size; returns total and each recordingId (pass it on to the get_* tools). To browse all recordings by date/folder, use list_recordings instead.