AssemblyAI n8n integrationAutomate AssemblyAI with n8n.

Sends a media file to AssemblyAI and returns a URL you can hand to a transcription. Use it whenever the audio is not already on a public address.

• File Path or Binary Data: either a path to the file to upload, or the name of the binary property carrying it, typically data when the file arrives from a download node.

Sends a chat completion request to a language model through the AssemblyAI gateway, and returns the model output. This is where summaries, chapter lists and extraction live now.

• Model: the model that answers, picked from a list that spans Claude, Gemini and GPT entries such as claude-sonnet-4-5-20250929 or gpt-5-mini.
• Transcript ID: optional. When set, the first literal {{ transcript }} tag found in the messages or in the Prompt is swapped for the transcript text before the model runs.
• Temperature and Max Tokens: how loose the wording gets, and how long the answer can run.

Runs a speech understanding task over a transcript that already exists, and returns the processed result rather than a new transcription.

• Transcript ID: the transcript to process, often {{ $json.id }} from the node that created it. Required, so an empty value stops the node.
• Task Type: translation, speaker_identification or custom_formatting.
• Target Languages: comma separated codes such as es,de.
• Speaker Type: name or role, and it needs the transcript to have had Speaker Labels enabled.

Starts a new transcription from an audio address. Everything that shapes the output is decided here, in an Additional Fields collection holding 48 options.

• Audio URL: the address of the audio to transcribe, usually the URL returned by Upload a file.
• Speaker Labels: identifies who is speaking, and several later options depend on it.
• Speech Models (Priority Order): a comma separated priority list such as universal-3-5-pro,universal-3-pro,universal-2. This is now the only way to choose a model.
• Key Terms: domain vocabulary to raise accuracy, up to 1000 terms for Slam-1 and 200 for Universal.

Removes a transcription from AssemblyAI. Nothing else in the workflow changes, so anything already written to a sheet or a message stays where it is.

• Transcript ID: the ID of the transcript to remove, commonly stored in the same row that received the text earlier.

Fetches a transcription by its ID and returns the full object, text and status included. Since Create a transcription returns before the audio has been processed, this is the node that tells you whether the work is done.

• Transcript ID: the ID of the transcript, usually carried from the create step as {{ $json.id }}.

Returns the transcript split into paragraphs instead of one block of text. Useful when the raw transcript is too long to hand straight to anything downstream.

• Transcript ID: the ID of the transcript whose paragraphs you want.

Returns the URL of the redacted audio file, the version where personal information has been masked in the sound itself.

• Transcript ID: the ID of the transcript the redacted audio belongs to. The file only exists if Redact PII Audio was enabled back at Create a transcription, so ask for it there first.

Returns the transcript cut into sentences, a finer grain than paragraphs and a better unit when each piece needs its own timing.

• Transcript ID: the ID of the transcript to break into sentences.

Returns ready to use subtitles for a finished transcription, in the caption format a player expects.

• Transcript ID: the ID of the transcript to get subtitles for.
• Format: srt or vtt, depending on what the player or the platform accepts.
• Characters Per Caption: the maximum length of a single caption, which is how you stop lines from filling the screen.

Lists the transcriptions on the account, with filters to narrow the batch before anything downstream touches it.

• Limit: the maximum number of transcripts returned, so a run stays predictable instead of pulling the whole history.
• Status: keeps only completed, error, processing or queued transcripts.
• After ID and Before ID: paging markers to walk through the account without repeating items.

Looks for specific words inside a transcript and returns where they appear, instead of handing back the whole text for something else to scan.

• Transcript ID: the ID of the transcript to search.
• Words to Search: either a comma separated string, or individual terms added one by one.
• Input Type: picks between the comma separated list and the individual terms.