Deepgram
AIGive agents Deepgram voice AI: transcribe recorded audio with diarization, summarization, and sentiment, generate speech with Aura voices, and analyze text for topics and intent. Also manages Deepgram projects, API keys, usage, and billing.
What This Integration Enables
Voice is the largest unstructured data source most operations teams have, and it is the one they act on least. This connector closes that gap. FlowRunner agents transcribe recorded audio with Deepgram's speech-to-text models, complete with diarization, summarization, and sentiment, generate speech with Aura voices, and analyze any text for topics and intent. The same connector manages the Deepgram account itself: projects, API keys, usage, and prepaid balances. That second half is where judgment lives. Transcription is cheap to retry; account operations are not. An agent can watch usage and balances all day, but revoking a key that a production service still depends on stops transcription everywhere that key is used, instantly. That decision goes through a [human-in-the-loop](/concepts/human-in-the-loop) gate with the usage evidence attached.
Without FlowRunner
With FlowRunner
Use Case Scenarios
The call QA pipeline that listens to everything
Recordings flow in from the phone system, or get pulled from [Google Drive](/integrations/google-drive-api) into file storage. Each one runs through Transcribe Audio File with diarization and summarization, then Analyze Text scores sentiment and intent. Routine calls file their transcripts and summaries quietly. Calls that show churn language, a compliance phrase, or escalating frustration post to [Slack](/integrations/slack) with the summary, the speaker breakdown, and a link to the moment in question. The CS lead reviews the handful that matter instead of sampling three calls at random.
Voice in, knowledge out
A weekly meeting recording becomes a speaker-labeled transcript, and the agent files the summary and decisions into [Notion](/integrations/notion) where the team actually looks. In the other direction, Convert Text to Speech turns approved announcement copy into an Aura-voiced audio file saved to FlowRunner file storage, ready for the IVR greeting or the product walkthrough. One connector covers both directions of the voice boundary.
Usage governance and key rotation with evidence
On a schedule, the agent pulls Get Usage Breakdown by endpoint and model, checks List Balances for remaining prepaid credit, and posts a burn-rate digest. When credit runs low or a key looks stale, the agent assembles the evidence: which keys exist from List API Keys, what each one actually processed from List Usage Requests. Stale-key candidates go to the platform owner for a decision before any revocation. New scoped keys get provisioned with Create API Key as services come online.
Human-in-Loop Highlight
Delete API Key is permanent and takes effect immediately, and its blast radius is invisible from inside Deepgram: the key that looks stale might be the one your IVR transcription quietly runs on. So the rotation agent never revokes on inference. It posts to the platform owner: "Key ending 7f2c, scoped to usage:write, last appears in request logs 41 days ago, tagged to the retired batch pipeline. Two other keys show active traffic. Approve revocation of 7f2c?" The owner confirms or corrects the story. Only then does the agent delete the key and log the rotation. Voice processing never goes dark because an agent guessed wrong about a dependency.
Agent Capabilities
19 actionsSpeech and Text Intelligence
4- Transcribe Audio from URL Transcribes pre-recorded audio hosted at a public URL using Deepgram's speech-to-text models. Used when recordings live in another system that serves URLs.
- Transcribe Audio File Transcribes an audio file from FlowRunner file storage, with diarization, summarization, and the full feature set. The workhorse of call QA pipelines.
- Convert Text to Speech Converts text into natural-sounding speech using Aura voices. Used for voiceovers, IVR prompts, and audio versions of approved copy.
- Analyze Text Analyzes written text with Deepgram Text Intelligence for topics, intent, and sentiment. The scoring step that decides which transcripts deserve a person's attention.
Projects
2- List Projects Retrieves the Deepgram projects the configured key can access.
- Get Project Retrieves a project's ID, name, and organization details.
API Keys
3- List API Keys Lists a project's keys with scopes, tags, and owning member. The inventory step in every rotation workflow.
- Create API Key Creates a scoped key with a comment. Used to provision per-service keys so usage stays attributable.
- Delete API Key Permanently deletes a key. The operation this page's human gate exists for.
Usage and Billing
7- Get Usage Summary Retrieves aggregated usage over a date range, resolved per day. The input to burn-rate digests.
- Get Usage Breakdown Retrieves usage grouped by endpoint, model, method, tag, or deployment. Used to attribute spend to workloads.
- List Usage Fields Lists the features, models, and methods used in a period.
- List Usage Requests Lists individual API requests, newest first. The evidence behind any claim that a key is stale.
- Get Usage Request Retrieves one request's path, response code, timing, and the key that made it.
- List Balances Lists outstanding prepaid credit balances with remaining amounts. The check that keeps a batch job from running the account dry.
- Get Balance Retrieves a specific balance's remaining amount and purchase order.
Models
3- List Models Lists all available speech-to-text models and text-to-speech voices with languages and versions.
- Get Model Retrieves one model's details by UUID.
- List Project Models Lists the models available to a specific project, including private or custom ones. Used so workflows request models the project can actually use.
Frequently Asked Questions
What can FlowRunner do with Deepgram?
FlowRunner agents can run Transcribe Audio from URL, Transcribe Audio File, and Convert Text to Speech in Deepgram, plus 16 more actions.
Does connecting Deepgram to FlowRunner require OAuth?
No. Deepgram connects to FlowRunner with an API key, no OAuth flow required.
Can Deepgram trigger a FlowRunner workflow automatically?
Deepgram doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger.
Start building with Deepgram
$100 in credits. No card required. Connect in minutes.