Deepgram
Transcribe audio and video by URL, analyse text for sentiment, topics and intent, generate speech, and watch the spend and request history behind it.
About
Deepgram ships in the w6w first-party pack. It declares 19 actions, 3 health checks, and the host runs its code in a sandbox that never sees the credential.
- App id
io.w6w.deepgram- Version
- 0.1.1
- Author
- w6w
- Licence
- MIT
- Categories
- AI & Machine Learning · Video & Streaming · Developer Tools
Overview
Deepgram turns audio and video into text and text into speech, and this app gives a workflow that pipeline without ever having to move the media itself — point it at a URL and Deepgram fetches the file directly, so a step can transcribe an hour of audio without holding a byte of it.
Beyond transcription, it can analyse text for sentiment, topics and intent, generate spoken audio from a script, and mint a short-lived token so a browser never needs the underlying API key. Because a long job or a text-to-speech request can outlive an ordinary request timeout, results are delivered to a callback rather than held open, and a companion action looks up what happened to a request that never seemed to finish.
The remaining actions turn Deepgram’s own account and spend data into something a workflow can act on: how many hours were billed and against which tags, how much prepaid credit is left, and who and what can access the project — useful for keeping an automated pipeline inside its budget rather than discovering the ceiling by hitting it.
Build with Deepgram
Three routes to the same 19 actions. The Workflow tab is generated from Deepgram's own manifest and carries its real ids, so it is copy-pasteable; the Code and CLI examples are the same call for any action on any app, so every app-specific value in them is a blank you fill in.
Transcribe audio from a URL
audio-transcribe Deepgram fetches the media itself, so nothing passes through the workflow. Give a callback URL for anything longer than a few minutes — a synchronous call will outlive its timeout.
List balances
balance-list Remaining pre-paid credit. Running out stops transcription rather than slowing it — and an invoiced account reports no balance at all, which is not the same as zero.
List invitations
invite-list Outstanding invitations — pending grants that `member-list` cannot see, and that Deepgram does not expire on its own.
Create an API key
key-create Mint a key for this project. The value is returned ONCE and never again — store it in the same run or delete it. Scopes are fixed at creation.
Delete an API key
key-delete Revoke a key immediately and permanently. Deleting the one this connection uses breaks it and every workflow on it — and nothing in the API says which id that is.
List API keys
key-list Which credentials reach this project and what each may do. Values are never returned — Deepgram shows a key once — so this is safe to schedule.
List members
member-list Who has access to this project. With `key-list` it is the whole access picture — the people and the machines reaching the same data.
List a member's scopes
member-scope-list What one person may do in this project. A member with `owner` can mint an owner key, so this and `key-list` only mean something read together.
List models
model-list Speech-to-text models and text-to-speech voices — Deepgram names both as models. Include outdated ones to explain an old request rather than to choose a new model.
Get this project
project-get The project this connection's key belongs to, including the contract shape that decides whether there is a pre-paid balance to read at all.
List projects
project-list The projects this key reaches — the unit of billing, keys, members and usage. Usually one, and the first thing to check when a usage figure looks wrong.
Get a request
request-get What happened to one request. The only place that distinguishes 'the callback never arrived' from 'Deepgram could not fetch the audio' from 'it never got here'.
List requests
request-list The request log, filterable to failures — how a workflow finds out that a tenth of last week's transcriptions never happened, which no individual step ever reported.
Generate speech from text
speech-generate Text to speech, delivered to a callback URL. That is required rather than optional — the synchronous form returns audio bytes, which a workflow step cannot usefully hold.
Analyse text
text-analyze Sentiment, topics, intents and summary over text that never was audio. Name your own categories rather than accepting the generic ones.
Grant a temporary token
token-grant Mint a short-lived JWT so a browser or device can reach Deepgram's streaming API without ever holding the API key. Sent as Bearer, not Token.
Get a usage breakdown
usage-breakdown-get Spend grouped by model, tag, endpoint or API key. A total that doubled is a fact; a total that doubled because one tag tripled is something you can act on.
List usage fields
usage-fields-list Which models, tags and features appeared in a period. Tags are not configured anywhere — they exist because a request carried one — so this is where a report finds them.
Get usage
usage-get Requests and audio hours over a date range. Deepgram bills by duration, so hours are the number that matters — not the request count.
A workflow step names the app and the action, and the editor fills in the
connection when you pick one. This is the Step shape from the
workflow spec, carrying Deepgram's real ids.
{
"manifestVersion": "2",
"name": "deepgram-example",
"steps": [
{
"id": "audio-transcribe",
"uses": {
"app": "io.w6w.deepgram",
"action": "audio-transcribe",
"connection": "conn_YOUR_CONNECTION_ID"
},
"with": {
"url": "<url>"
}
}
]
}Here are some of the things you can do
-
Transcribe audio from a URL
performaudio-transcribe -
List balances
readbalance-list -
List invitations
readinvite-list -
Create an API key
performkey-create -
List API keys
readkey-list
+14 more actions available
Every app-specific value here is a blank you have to fill in. An
app action is reached through the connection that authenticates it, so the
address is a connection id, not the app id — and connections belong to your account,
so a public page cannot know yours. Create one for Deepgram, then fill in
the three blanks: conn_YOUR_CONNECTION_ID, the action key, and the
parameters that action declares. The call itself is real — the shape is transcribed
from the studio's own snippet builder, which prints the same kind of blanks — but
nothing in it is specific to Deepgram. The Workflow tab is where this app's
real ids are.
npm install @w6w/sdkyarn add @w6w/sdkpnpm add @w6w/sdkdeno add npm:@w6w/sdkimport { W6wClient, isActionRun } from "@w6w/sdk";
// Reads W6W_BASE_URL and W6W_TOKEN from the environment when omitted.
const client = new W6wClient();
const envelope = await client.run({
urn: "conn_YOUR_CONNECTION_ID",
action: "audio-transcribe",
payload: {
url: "<value>",
// callbackUrl: "<value>",
// model: "<value>",
// language: "<value>",
// detectLanguage: "<value>",
// smartFormat: "<value>",
// punctuate: "<value>",
// paragraphs: "<value>",
// diarize: "<value>",
// keyterm: "<value>",
// redact: "<value>",
// mipOptOut: "<value>",
// summarize: "<value>",
// topics: "<value>",
// sentiment: "<value>",
// intents: "<value>",
// detectEntities: "<value>",
// multichannel: "<value>",
// tag: "<value>",
},
});
if (isActionRun(envelope)) console.log(envelope.value);npm install -g @w6w/cliw6w run conn_YOUR_CONNECTION_ID --action audio-transcribe --payload '{"url":"<value>"}'Give an AI agent Deepgram — without giving it Deepgram's credentials. One MCP endpoint exposes every app, function and workflow the caller is entitled to, as tools it can discover and run. Access is granted per team while we onboard.
One tool call{
"name": "w6w_invoke",
"arguments": {
"ref": "app:io.w6w.deepgram#audio-transcribe",
"input": {
"url": "<url>"
}
}
}
Every tool names its target with a single ref. The
app: form above doesn't name a connection at all — the
host resolves which of the caller's Deepgram connections to sign
with, and refuses rather than guesses when the answer is ambiguous.
What the agent gets
Credentials it can't read
The token is attached host-side, at the moment of the call. It is never a tool argument, never in the model's context, and never in a transcript — so a prompt injection has nothing to exfiltrate.
A tool surface scoped to the caller
Tools are derived per end user from what that person has actually connected and is entitled to — not one shared bot identity carrying the union of everyone's access.
A durable workflow in one call
Multi-step work runs on the workflow engine and returns a run handle the agent can poll — retries, branching and state survive the conversation that started them.
Health-aware discovery
Deepgram's declared health checks are on the surface too, so an agent can tell "the vendor is down" from "your credential expired" before it burns a retry on either.
The MCP surface is part of the hosted platform. Deepgram itself is MIT, and the runtime that executes it is source-available (FSL).
Health checks
Deepgram declares its own checks, so its health is a property of the app rather than something the host guesses at.
Deepgram platform status
The batch, TTS, usage and management surfaces this app calls. Streaming and Voice Agent are WebSocket surfaces it cannot reach, so their outages do not count.
Credit remaining
The project's pre-paid balance — genuine headroom rather than a rate limit, because running out does not slow transcription down, it stops it.