AI Providers

Every agent runs on three engines that you choose and plug in yourself: an LLM (the brain that decides what to say), Speech-to-Text / ASR (the ears that hear the caller), and Text-to-Speech / TTS (the voice it speaks with). InfraVoice hosts no models of its own — you bring your own provider keys, and you can mix and match any brain, ears, and voice per agent.

You set all of this up in the dashboard at your-infravoice-host — there's nothing to install and no code to write.

All three (LLM + Speech-to-Text + Text-to-Speech) must be set before an agent can take a live call. Because you supply the keys, you're billed directly by each provider for what your agent uses.

---

The three engines

There's also an optional fourth slot, the Analysis LLM, used only for post-call work (summaries, QA scoring, checklist tracking). Leave it empty and the agent's main LLM is used instead — pick a separate, often cheaper, model here if you'd rather run analysis on something lighter than your live voice model.

---

How to add providers

Everything happens on one screen: open an agent from the Agents page and switch to its Providers tab.

1. In the panel for the engine you want (LLM, Speech-to-Text, or Text-to-Speech), choose a Provider from the dropdown.
2. Paste the provider's API Key. Depending on the provider, you may also fill in a Base URL, a Model, and — for Text-to-Speech only — a Voice ID. The form shows only the fields that provider needs and pre-fills sensible defaults (for example a recommended model name), so most of the time you only paste a key.
3. Click Save LLM, Save ASR, or Save TTS. Your key is validated with the provider before it's stored, so you'll know right away if it's wrong.

That's it — repeat for all three engines and your agent has a brain, ears, and a voice.

Tip: Some voices offer extra tuning. Text-to-Speech panels for providers with preset voices let you pick from a list (or choose Custom… to paste your own voice ID), and a few providers expose an Advanced Options panel for things like speaking rate and quality.

Remember to publish. Saving a provider updates a draft. Use the version banner at the top of the editor to Publish the new version so live calls start using it.

---

Reuse a key across agents — Credentials

Rather than pasting the same key into every agent, save it once and reuse it everywhere. Under Build → Credentials you'll find Provider Credentials:

1. Click Add credential.
2. Choose the Type — LLM, Speech-to-Text, or Text-to-Speech.
3. Pick the Provider, give it a Name, and paste the API Key (plus optional Base URL, Model, or Voice ID).
4. Save. The key is validated before it's stored, encrypted at rest, scoped to your workspace, and never shown in full again.

Back on an agent's Providers tab, a saved credential appears in a picker labelled (reuse across agents) — select it instead of typing a key. Prefer to enter a key inline? You can tick Save to workspace credentials for reuse while saving so it's available to your other agents next time.

---

Supported provider families

You're not locked into any one vendor — swap providers per agent at any time. The dashboard's dropdowns list every option; here's the landscape at a glance.

LLM (the brain)

Speech-to-Text / ASR (the ears)

Text-to-Speech / TTS (the voice)

---

Choosing a good combination

Any brain, ears, and voice will work together — but a few pairings are worth knowing:

When you've saved all three, head to Voice Test to hear your agent live in the browser before you put it on a phone number.