Temi
AISubmit audio or video to Temi by Rev for machine transcription and get back plain text or structured JSON with speaker labels and word timings. Agents make recorded conversation searchable and quotable.
What This Integration Enables
Temi is Rev's machine transcription service, and it is refreshingly honest about being a machine. That honesty is exactly what makes it work inside a flow. It hands back two distinct things: Machine Generated, the untouched speech recognition output, and Latest, which includes any corrections a person has made in the Temi editor. Most transcription tools blur that line. Temi keeps it, and the gap between those two versions is precisely where a human belongs.
An agent submits a public URL or uploads a file from FlowRunner file storage, polls the job until it reports transcribed, and then pulls the result as plain text or as structured JSON carrying speakers, monologues, and word level timestamps. It can render the transcript to PDF or Word and store it for an email attachment or a matter file, and it can mint a share link so a colleague without a Temi account reads and corrects the transcript in their browser. Billing is per minute of media rounded up, so a flow that checks the account balance before submitting a two hour recording is not being cautious for its own sake. Temi's own API is the source here, not the Rev.ai API, and this connector is built and verified against it.
Without FlowRunner
With FlowRunner
Use Case Scenarios
Client calls that become a searchable record
Sales and services calls are recorded and dropped on a shared drive. Overnight, the agent uploads each one with Upload Media File, polls Get Job, and pulls structured JSON from Get Transcript. It writes the text and the speaker segments into the account record so anyone can search what was actually said on a call from four months ago. Where the machine output disagrees with itself about who was speaking, the agent does not guess: it shares the transcript back to a participant and waits.
Transcripts filed as documents
For interviews, hearings, and intake calls, the text is not the deliverable, the document is. After a job reaches transcribed, the agent calls Save Transcript To File to produce a PDF or Word file, stores it in FlowRunner file storage, and attaches the durable URL to the matter or case record. The connector checks the downloaded bytes against the expected document signature, so an error response never gets filed as if it were a transcript. That check matters more than it sounds when the file is going into a record somebody will rely on.
Meetings turned into commitments
A recorded internal meeting is transcribed, and the word timings let the agent link each extracted item back to the moment it was said. It summarizes the transcript with OpenAI, drafts the action items into Notion with a timestamp link beside each one, and posts the draft list into Slack for the meeting owner to confirm before anything gets assigned. Nobody is asked to trust a summary of a transcript nobody read.
Human-in-Loop Highlight
The gate on this connector is about what counts as the record. A machine transcript is an approximation, and speech recognition fails in specific and predictable places: proper nouns, numbers, crosstalk, and the boundary where one speaker takes over from another mid sentence. When an agent copies that output into a client record, a matter file, or a quote attributed to a named person, the approximation quietly becomes the thing everyone treats as true.
So the agent refuses to promote Machine Generated output into a system of record on its own. When it flags a passage, it calls Share Transcript and sends the link to a participant with a direct message: "The transcript of the Feb 12 call with Harrison has three passages where the speaker changes mid sentence, at 04:12, 11:47, and 22:03, and the client's surname is spelled two different ways. Open the editor and correct them, and I will file the corrected version." The participant fixes it in the browser without needing a Temi account of their own. The agent then re-reads the transcript with Latest, which now carries those edits, and files that. The machine did the transcription. A person who was in the room decided what the record says.
Agent Capabilities
9 actionsTranscription Jobs
5- Submit Media URL Queue a publicly reachable audio or video URL for transcription, returning a job in the in_progress state. Supports an optional per job callback URL and metadata. Transcription is asynchronous, so poll Get Job until the status is transcribed.
- Upload Media File Upload an audio or video file to Temi as a multipart request and create a transcription job. Use this when the media lives in FlowRunner file storage or behind a private link that Temi cannot fetch itself.
- Get Job Retrieve a single job including its status, media duration, name, metadata, the Temi web editor URL, and any failure detail. This is the polling companion to both submission actions.
- List Jobs Return transcription jobs for the account, newest first, with keyset pagination via the ID of the last job on the previous page. Used to reconcile a batch or to sweep for jobs that failed.
- Delete Job Permanently delete a job and its transcript from Temi. A job still being transcribed cannot be deleted and the request is rejected until transcription finishes or fails. There is no recovery, so treat retention cleanup as its own reviewed workflow.
Transcripts
3- Get Transcript Retrieve the transcript of a transcribed job as plain text or as structured JSON with speakers, monologues, and word level timestamps. Choose Latest to include corrections made in the Temi editor, or Machine Generated for the untouched recognition output. The choice between those two is a decision about what your record says.
- Save Transcript To File Download a transcript as a PDF or Word document and store it in FlowRunner file storage, returning a durable URL for later steps such as an email attachment or a case file. The downloaded bytes are validated against the expected document signature so an error response is never saved as a document.
- Share Transcript Create a shareable Temi editor URL for a transcribed job so a colleague can read and correct the transcript in the browser without a Temi account of their own. This is the action that turns a review request into something a participant can actually complete.
Account
1- Get Account Return the account balance and the email the API key belongs to. Used to verify credentials at the start of a flow, or to confirm available funds before submitting a long recording, since Temi bills per minute of media rounded up.
Triggers
1 triggersEvent Triggers
1- On Transcription Ready Fires once for every job that reaches the transcribed state. Temi's callback URL is a per job field on the submit call rather than a webhook subscription API, so this trigger polls the job list and emits only jobs it has not reported before, at a configurable interval with a thirty second minimum. Use it to start the downstream work the moment a transcript exists: file the document, extract the commitments, route the flagged passages for correction.
Frequently Asked Questions
What can FlowRunner do with Temi?
FlowRunner agents can run Submit Media URL, Upload Media File, and Get Job in Temi, plus 6 more actions.
Does connecting Temi to FlowRunner require OAuth?
No. Temi connects to FlowRunner with an API key, no OAuth flow required.
Can Temi trigger a FlowRunner workflow automatically?
Yes. Temi supports 1 trigger that can start a FlowRunner workflow automatically.
Start building with Temi
$100 in credits. No card required. Connect in minutes.