> ## Documentation Index
> Fetch the complete documentation index at: https://docs.voiceinfra.ai/docs/llms.txt
> Use this file to discover all available pages before exploring further.
# Connect Any AI Provider to Your VoiceInfra Voice Agents
> VoiceInfra supports OpenAI, Anthropic, Google Gemini, Groq, and Mistral. Switch providers anytime, run A/B tests, and cut costs with smart routing.
VoiceInfra lets you connect five leading AI providers in one platform and assign any of them to any agent or workflow node — independently. You're never locked into a single vendor. If a better model launches, you switch from a dropdown. If a provider has an outage, your backup model takes over automatically. If you want to cut costs on simple FAQ calls, route them to a lightweight model and reserve your premium model for high-stakes conversations. The same agent configuration, prompts, and integrations work regardless of which model is underneath.
## Supported Providers
| Provider | Available Models | Best For |
| ------------- | ----------------------------------------- | ------------------------------------------------------ |
| **OpenAI** | GPT-4x, GPT-5x, GPT-Realtime | General purpose, high output quality, broad capability |
| **Anthropic** | Claude Sonnet and Claude Haiku all models | Complex reasoning, nuanced multi-turn conversation |
| **Google** | Gemini 3x, Gemini 2x, Gemini Pro | Multilingual calls, multimodal context |
| **Groq** | OpenAI GPT OSS, Llama, Qwen, Mixtral | Speed-critical calls, ultra-low inference latency |
| **Mistral** | Mistral Large, Medium and Small | Cost-efficient operation, European data residency |
## Smart Cost Routing
Not every call needs the same model. A caller asking for your business hours needs a fast, cheap response. A caller troubleshooting a complex technical issue benefits from deeper reasoning. Smart cost routing lets you match the model to the moment.
**Example cost comparison (approximate per-minute LLM costs):**
| Model | Approximate cost | Best used for |
| ---------------- | ---------------- | -------------------------------------- |
| GPT-4.1 Nano | \~\$0.004 / min | FAQs, routing, simple data collection |
| GPT-4.1 Mini | \~\$0.012 / min | Standard support, lead qualification |
| GPT-4o | \~\$0.06 / min | Complex reasoning, sales conversations |
| Claude 3 Haiku | \~\$0.005 / min | Fast, nuanced responses at low cost |
| Groq Llama 3 70B | \~\$0.008 / min | Speed-first, high-volume deployments |
Costs vary by provider pricing and are passed through at provider rates when you use your own API keys. Check each provider's current pricing page for up-to-date figures. These estimates are for LLM usage only; voice (TTS/STT) costs are separate.
By routing simple calls to lightweight models and complex calls to premium models, teams typically reduce their blended AI cost by 40–60% compared to running every call on a single premium model.
## Setting Up a Provider
Open the VoiceInfra dashboard and navigate to **Settings** in the left sidebar, then select **AI Providers**.
Click the provider you want to add (Anthropic, OpenAI, Google, Groq, or Mistral). Paste your API key from that provider's dashboard. VoiceInfra validates the key immediately.
Once saved, the provider and all of its supported models appear in the model picker across every agent and workflow node in your account. No restart or republish required.
In the no-code builder, select your model under the **Model & Voice** tab. In the workflow builder, each Conversation node has its own model selector — so different nodes in the same workflow can use different providers.
## Zero Vendor Lock-In
VoiceInfra is designed so your agents are portable across providers. Your configurations, prompts, behavioral rules, knowledge bases, and tool integrations are all stored independently of the AI provider. Switching models never touches any of that.
* **Switch instantly** — change the model for any agent or node from a single dropdown. The change applies to new calls immediately.
* **Side-by-side testing** — run two agents with identical configurations but different models. Compare transcript quality, response accuracy, and cost in the call logs to find the right model for each use case.
## Bring Your Own API Keys
When you connect a provider using your own API key, AI inference costs are billed directly by the provider at their standard rates. VoiceInfra does not mark up pass-through AI costs. This gives you full visibility into what each call costs and lets you optimize at the provider level — including taking advantage of volume discounts, committed use agreements, or enterprise pricing you've negotiated directly.
If you prefer not to manage individual provider accounts, VoiceInfra also offers managed access to supported models through platform credits.
**Provider-specific prompt behavior:** Different models have distinct strengths you can use to your advantage. Claude 3.5 Sonnet handles long, nuanced conversations especially well and is less likely to veer off-topic on complex calls. Groq's Llama inference is optimized for speed — ideal when response latency matters more than depth. Gemini 1.5 Flash balances speed and multilingual accuracy for international deployments. When switching providers, review your system prompts to make sure tone and instruction style are aligned with the new model's behavior.
***
## Frequently Asked Questions
No, API keys are optional. VoiceInfra provides managed access to all supported models through platform credits — no provider account needed to get started. If you bring your own API keys, AI costs are passed through at the provider's standard rates, which gives you more cost control and lets you use any pricing agreements you have directly with the provider.
Yes, VoiceInfra supports any custom model with OpenAI-compatible API. If needed we will create a custom plugin for it.