AssemblyAI - Atako

Let your agents transcribe audio and video from public URLs with AssemblyAI (speaker labels, language detection, key phrases, entities, topics, sentiment, PII redaction), then read transcripts, sentences, paragraphs, subtitles and word matches.

Connection

Authentication: API key.
Required settings:
- Region — “api” for the default US endpoint, or “api.eu” for EU data residency (api.eu.assemblyai.com).

Sign in to the AssemblyAI dashboard (assemblyai.com/dashboard) → in the sidebar, Workspace → Manage → API Keys → Create New API Key. Enter a descriptive name, click Create and copy the key. Keys have no scopes: a key gives access to its project’s transcripts. For EU data residency, set the region to “api.eu”.See AssemblyAI’s documentation.

Read actions (7)

Action	Description
`get_redacted_audio`	Get the status and download URL of the PII-redacted audio copy of a transcript created with redact_pii_audio true. The URL is temporary (about 24 hours).
`get_subtitles`	Export a completed transcript as subtitles. Arguments: transcript_id (string), subtitle_format (“srt” or “vtt”), chars_per_caption (optional integer — maximum characters per caption). Returns the subtitle file as plain text.
`get_transcript`	Get a transcript by id: status (“queued”, “processing”, “completed”, “error”), error message, full text, words, utterances (with speaker_labels), detected language, and the results of the models enabled at creation — key phrases (auto_highlights_result), entities, topics (iab_categories_result), sentiment_analysis_results. Find ids with list_transcripts.
`get_transcript_paragraphs`	Get a completed transcript split into paragraphs, each with its text, start/end timestamps (ms), confidence and words.
`get_transcript_sentences`	Get a completed transcript split into sentences, each with its text, start/end timestamps (ms), confidence, words and speaker.
`list_transcripts`	List the transcripts of the key’s project, newest first. Optional filters: limit (1-200, default 10), status (“queued”, “processing”, “completed”, “error”), created_on (YYYY-MM-DD), before_id / after_id (transcript ids, for pagination — page_details.prev_url points to older transcripts), throttled_only (boolean). Each item gives id, status, created, completed and audio_url.
`search_transcript_words`	Search a completed transcript for words or phrases. Arguments: transcript_id (string), words (array of strings, at least one). Returns the total count and, for each match, the text, count, timestamps and word indexes.

Write actions (1)

Action	Description
`create_transcript`	Queue the transcription of an audio or video file reachable at a public URL (no file upload). Returns the transcript object with its id and status “queued”; poll get_transcript until status is “completed” or “error”. Arguments: audio_url (string, required — public https URL of the media); language_code (string, e.g. “en”, “fr”, “en_us” — cannot be combined with language_detection); language_detection (boolean — detect the spoken language automatically); speech_models (array of strings among “universal-3-5-pro”, “universal-2”, in priority order); punctuate (boolean); format_text (boolean — casing and formatting; required true for redact_pii); speaker_labels (boolean — speaker diarization, needs punctuate); speakers_expected (positive integer, needs speaker_labels); disfluencies (boolean — keep filler words); filter_profanity (boolean); multichannel (boolean); audio_start_from and audio_end_at (integers, milliseconds); keyterms_prompt (array of strings — domain terms, max 6 words each); prompt (string — context for the universal-3-5-pro model); custom_spelling (array of objects { from: array of strings, to: string }); auto_highlights (boolean — key phrases); entity_detection (boolean); iab_categories (boolean — topic detection); sentiment_analysis (boolean, needs punctuate); redact_pii (boolean, needs format_text true); redact_pii_policies (array of policy strings such as “person_name”, “email_address”, “phone_number”); redact_pii_sub (string “entity_name” or “hash”); redact_pii_audio (boolean — also produce a redacted audio copy, needs redact_pii; read it with get_redacted_audio).

Permissions

Every action above must be explicitly granted to an agent before it can be used. See Permissions for the grant model and Security for how credentials are protected.

Last reviewed against the provider API: September 2026.