Ringg.AI integration with Truefoundry AI Gateway

As voice agents move from demos into production, teams hit a familiar wall. Speech is the hardest part of the stack to get right, especially in markets where people do not speak one clean language at a time. A customer in India might open a call in Hindi, switch to English mid-sentence, and rattle off an order number in a code-mixed blur. Most speech-to-text models trip over exactly this. And even once you find one that works, you are left wiring yet another provider into your gateway, managing another API key, and hoping your logging and access controls stretch to cover it.

The question every voice team eventually asks: how do you plug a specialized speech model into your platform without turning it into a one-off engineering project?

At TrueFoundry, our goal is to make every model, no matter how specialized, feel like a first-class citizen of your gateway. That is why we are excited to bring Ringg.AI's Parrot speech-to-text model to the TrueFoundry AI Gateway. With a single Custom Endpoint configuration, your teams can call Parrot the same way they call any other model on the gateway, with the same authentication, the same access controls, and the same centralized visibility.

TrueFoundry AI Gateway as a Central Control Plane

The TrueFoundry AI Gateway is how developers and platform teams manage, monitor, and scale their AI applications from one place. It gives you unified access to hundreds of models, smart routing, virtual keys, and full observability, all behind a single API.

As AI stacks grow, the real challenge is not reaching a model. It is the sprawl that comes after. Every new provider brings its own SDK, its own keys, and its own way of doing things. The Gateway brings order to that. It sits between your applications and your models as a single control point, so adding a new capability does not mean scattering credentials and custom code across your codebase. That is what makes it the natural home for a specialized model like Parrot: you get the speech model you need without giving up the consistency you already rely on.

Ringg.AI Parrot: Speech-to-Text Built for Real Voice

Parrot is Ringg.AI's speech-to-text model, purpose-built for real-time voice agents. Where general-purpose transcription models are trained mostly on clean, single-language audio, Parrot is designed for how people actually talk, with strong support for Hindi, English, and the code-mixed speech that dominates real conversations.

That focus matters. A voice agent is only as good as its ears. If transcription drops a word or mangles a switch between languages, every downstream step, from intent detection to response generation, inherits the error. By handling messy real-world speech accurately and in real time, Parrot gives voice agents a reliable foundation to build on. It is the difference between a demo that works in a quiet room and an agent that holds up on a live customer call.

Parrot's strengths show up in the details that trip up general models. It normalizes Devanagari output so the transcript is clean enough for an LLM to reason over directly. It handles alphanumeric strings, like order numbers and IDs, more accurately than typical market options, and does better at recognizing proper nouns and named entities. And it does all of this while ranking among the fastest speech-to-text models available.
For example, Parrot handles an utterance like "अभी तो I am taking hydroxychloroquine sulfate" cleanly, correctly transcribing the Hindi-English switch, the normalized Devanagari, and the drug name in one pass. 

Better Together: One Endpoint, Full Control

Bringing Parrot onto the TrueFoundry AI Gateway means you no longer treat speech as a separate, bolted-on service. Parrot becomes just another endpoint on your gateway, governed by the same rules as everything else.

The benefit is a clean separation of concerns. Your Ringg.AI API key lives inside the gateway and is sent upstream from gateway to Ringg on every request. Your applications never see it. Your client only ever holds a TrueFoundry API key and talks to the gateway. This means you can rotate credentials, manage who is allowed to use Parrot, and audit usage from one dashboard, without touching application code. You configure Parrot once as a Custom Endpoint, set access control for the right users and teams, and it is available across your organization with the governance you already expect.

How the Ringg.AI and TrueFoundry Integration Works

Parrot connects to the Gateway through Custom Endpoints, TrueFoundry's mechanism for putting any HTTPS API behind the gateway's proxy path. The diagram below shows the end-to-end flow.

‍

The layered flow works like this:

1.  Your voice application sends an audio file to the TrueFoundry AI Gateway using a multipart form request, authenticated with only a TrueFoundry API key.

2.  The Gateway forwards the request upstream to Ringg.AI Parrot, attaching the x-api-key header it holds internally. Optional parameters such as language and enable_cap_punc are passed along.

3.  Parrot transcribes the speech and returns the transcript as JSON back to the Gateway.

4.  Optionally, the transcript can be routed onward to any LLM already on the gateway, so a single control plane covers both the speech and reasoning steps of your voice agent.

5.  The Gateway returns the transcript or downstream response to your application, with the full request logged for visibility.

Because the integration rides on Custom Endpoints, there are no sidecars, no SDK swaps, and no per-model credential handling in your application. The gateway proxies the request, injects the upstream key, and records the call.

Configuration in Three Steps

Setup is closer to flipping a switch than an engineering project. In the TrueFoundry dashboard, go to AI Gateway → Models → Custom Endpoints and:

1.  Create a Custom Endpoint provider account. Give it a unique name, which becomes the providerAccountName in your call URL. Leave Header Auth disabled.

2.  Add a Ringg.AI STT endpoint. Set the display name (this becomes endpointName), set the Base URL to https://prod-api.ringg.ai/stt/v1/transcriptions, enable Custom Headers, and add x-api-key with your Ringg.AI key.

3.  Set access control for the users and teams who should use this endpoint, then save.

Once saved, you call Parrot through the gateway proxy path: {GATEWAY_BASE_URL}/proxy-api/{providerAccountName}/{endpointName}. Note the current support scope: only HTTPS and non-streaming requests are supported today, with more coverage added on customer request.

Get Started with Real-Time Voice AI

Voice AI does not need a separate integration path for every model. With Ringg.AI Parrot on the TrueFoundry AI Gateway, you get speech-to-text built for real, code-mixed conversation, wired into the same unified API, access controls, and observability you already use for the rest of your stack. To learn more, see the Ringg.AI integration reference in the TrueFoundry docs and the Parrot STT model page to start building.