AssemblyAI
AITranscribe recorded audio and video with AssemblyAI. Agents upload media, run speech-to-text with speaker labels, PII redaction, sentiment, chapters, and entity detection, export SRT or VTT subtitles, and run LLM prompts over finished transcripts.
What This Integration Enables
Call recordings are the least read documents in any company. AssemblyAI turns them into structured data an agent can actually work with: transcripts with per-word timestamps, speaker labels, sentiment, entities, chapters, and PII redaction applied at transcription time. FlowRunner agents upload media, run transcription jobs asynchronously or blocking, pull results as sentences, paragraphs, or SRT and VTT subtitle files, and then interrogate the finished transcript with LLM prompts through the AssemblyAI LLM Gateway. The pattern that matters is the second half. Transcription alone gives you text. The Ask Questions About Transcript and Run LLM Task operations give you answers with the source pinned to the recording, which is what makes full-coverage call QA, meeting documentation, and compliance search jobs an agent can own, with [human-in-the-loop](/concepts/human-in-the-loop) review reserved for the calls that deserve it.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Call QA that reviews everything, not a sample
Every recording from the support queue is uploaded with Upload Audio and submitted through Transcribe Audio with speaker diarization and sentiment analysis. When Get Transcript reports completion, Ask Questions About Transcript scores each call against the team's rubric: was the disclosure read, was the refund policy stated correctly, did the customer threaten churn. Answers come back per question, and the model replies "Not found in the transcript" rather than guessing. Calls that fail a rubric item post to [Slack](/integrations/slack) with Word Search timestamps, so the QA lead jumps straight to the moment instead of scrubbing through audio.
Meetings that document themselves
When a recording lands in [Dropbox](/integrations/dropbox-service), the agent grabs a temporary link and runs Transcribe and Wait with speaker labels. Get Paragraphs returns readable, speaker-attributed text that flows into a new document via [Google Docs](/integrations/google-docs), while Summarize Transcript produces the outline and Run LLM Task extracts action items with owners. The summary posts to the project channel the same hour the meeting ends, and the full transcript is one search away.
Publishing pipeline with captions and redaction
A podcast or video team drops finished episodes into storage. The agent transcribes each one, then Get Subtitles renders SRT captions, saves them to FlowRunner file storage, and hands the caption file to the editor through Dropbox. For customer call archives, the same pipeline runs with PII redaction enabled, so the stored transcript carries no card numbers or health details. Once the retention owner confirms the redacted copy is complete, Delete Transcript purges the original from AssemblyAI.
Human-in-Loop Highlight
Delete Transcript is permanent. It removes the transcript text, the per-word timings, and every Audio Intelligence result attached to the job, and it cannot be undone. In a retention workflow that is exactly the point, which is why the purge is the one step the agent never takes alone. The agent assembles the evidence: here is the redacted copy in the system of record, here is the retention window, here are the three transcripts tied to accounts with open disputes. The compliance owner releases the purge on the clean ones and holds the rest. Redaction runs at machine speed; the decision to destroy a record of what a customer was told belongs to a person.
Agent Capabilities
13 actionsAudio Files
1- Upload Audio Streams a local or remote audio or video file to AssemblyAI and returns a private upload URL for transcription. Use it when the source file is not publicly reachable.
Transcription
2- Transcribe Audio Submits a recording for asynchronous transcription and returns immediately with a queued transcript. Enables Audio Intelligence features in the same request: speaker diarization, language detection, PII redaction, chapters, highlights, sentiment, entities, topics, and summarization.
- Transcribe and Wait Submits a recording and blocks until the transcript reaches a terminal status, polling every 5 seconds up to a configurable cap. The right choice for short recordings inside a flow that needs the text now.
Transcript Management
3- Get Transcript Retrieves a transcript by ID with its status and, once completed, the full text, per-word timings, confidence scores, and every enabled Audio Intelligence result. The polling step behind asynchronous jobs.
- List Transcripts Lists the account's transcripts newest first with status and timestamps. Used for auditing transcription history or recovering a recent job's ID.
- Delete Transcript Permanently removes a transcript and all its results. The operation this page's human gate exists for.
Transcript Results
4- Get Sentences Returns a completed transcript split into sentences with timestamps, confidence, and speaker labels. Used for navigable transcript views and audio alignment.
- Get Paragraphs Returns the transcript as readable paragraphs with the same detail. The best shape for publishing or feeding chunks to an LLM.
- Get Subtitles Generates SRT or WebVTT captions from a completed transcript, optionally capped at a character width and saved to FlowRunner file storage as a downloadable file.
- Word Search Searches a transcript for words or phrases and returns occurrence counts with millisecond timestamp ranges. The step that turns a compliance term into a jump-to moment.
LLM Analysis
3- Run LLM Task Runs a custom LLM prompt over one or more completed transcripts, with the speaker-labelled text prepended automatically. Used for extraction, classification, and drafting follow-ups grounded in what was said.
- Summarize Transcript Produces an LLM summary of one or more transcripts with a selectable shape, from a single headline to a detailed outline. Re-runnable with different instructions after transcription.
- Ask Questions About Transcript Answers a list of questions strictly from the transcript, one answer per question in the format you specify, replying "Not found in the transcript" when the information is absent. The backbone of structured call QA.
Frequently Asked Questions
What can FlowRunner do with AssemblyAI?
FlowRunner agents can run Upload Audio, Transcribe Audio, and Transcribe and Wait in AssemblyAI, plus 10 more actions.
Does connecting AssemblyAI to FlowRunner require OAuth?
No. AssemblyAI connects to FlowRunner with an API key, no OAuth flow required.
Can AssemblyAI trigger a FlowRunner workflow automatically?
AssemblyAI doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with AssemblyAI
$100 in credits. No card required. Connect in minutes.