Deepgram is a voice-AI platform providing speech-to-text (Nova-3), text-to-speech (Aura-2), and text intelligence (summarization, sentiment, topics and intents). Transcribe audio from a URL or file, generate speech from text, analyze text and manage projects, requests and models.
Changelog
1 version
v1.0.0
Latest
- Initial version with Transcribe Audio, Text to Speech, Analyze Text, List Models, Get Model, List Projects, Get Project, Find Requests, Get Request and Make API Call action components, and New Request, Transcription Completed and Failed Request polling triggers.
- Transcribe Audio submits the job to Deepgram and returns the request ID immediately on "out" without holding a worker, then emits the finished transcript on "done" when Deepgram calls the component's own webhook back.
- Transcribe Audio echoes the job's input (Audio URL / File, plus an optional Correlation ID of your own) on both "out" and "done", so parallel transcriptions stay tellable apart: one component instance has a single callback URL and jobs finish in whatever order Deepgram completes them.
Authentication
| Type | Details |
|---|---|
| API Key |
API Key (password)
Region (select)
Deepgram API region. Your API key works against every region — this only selects the base URL. Pick Custom to target a dedicated or self-hosted endpoint.
Custom Host (text)
Only used when Region is set to Custom. Full base URL or host of a dedicated / self-hosted Deepgram endpoint, e.g.
https://abc123.us.api.deepgram.com or http://10.0.1.100:8080. |
Triggers
3
Failed Request
Triggers when a request fails in the selected Deepgram project (e.g. 402 out of credits, 429 rate limit, 504 timeout). Useful for surfacing failures in async pipelines that would otherwise be invisible.
New Request
Triggers when a new API request is logged for the selected Deepgram project. Polls the project request log and deduplicates by request ID.
Transcription Completed
Triggers when a transcription (/v1/listen) request succeeds in the selected Deepgram project. Optionally re-fetches the full stored result for each new request. Use it to react to transcriptions started outside this flow; a transcription started by Transcribe Audio reports back on that component's own "done" port.
Actions
10
Analyze Text
Runs Deepgram Text Intelligence (summarization, sentiment, topics and intents) over text or a text URL. Enable at least one analysis feature.
Find Requests
Finds API requests logged for a Deepgram project, filtered by date range and status. Returns up to 1000 request-log entries (a log, not a transcript store). Retrieve the stored result of a request with the Get Request component.
Get Request
Retrieves a single request-log entry and its stored result by request ID. Useful for re-reading the stored result of a past request, or for inspecting a request surfaced by the New Request, Transcription Completed or Failed Request triggers.
List Models
Lists the speech-to-text and text-to-speech models available on Deepgram, including languages and metadata. Returns an unpaginated array. Powers the model / voice dropdowns on other components.
List Projects
Lists all Deepgram projects visible to the API key. Returns an unpaginated array of projects.
Make API Call
Performs an arbitrary authorized API call to the Deepgram API. The path is resolved against the base URL of the selected region.
Text to Speech
Generates speech audio from text using Deepgram Aura-2 and stores the result in File Storage, returning a file ID. Text is limited to 2000 characters per request.
Transcribe Audio
Transcribes pre-recorded audio. Submits the job to Deepgram and returns the request ID immediately on the "out" port without holding a worker, then emits the finished transcript on the "done" port when Deepgram delivers it. Accepts an audio URL or a file (max 2 GB), with optional diarization, summarization, sentiment, topics and intents.