# Supafone Labs **An S2S routing hub with Supervisor. Choose the voice model. Keep the agent.** Supafone gives you one API for **Ultravox, OpenAI, Gemini, Grok and Smallest AI Hydra**, with a separate Supervisor that observes the call and guides the speaking agent. Choose a model, build your agent, then switch models without rebuilding its instructions, tools or call stages. Start with **one Supafone API key**, or bring your own keys for the speaking model and the Supervisor. You can choose who supplies each key independently. [Build your first S2S agent](quickstart.md) · [Open Agent Factory](https://app.supafone.ai/app/agents) · [Install Python or TypeScript](sdk-installation.md) · [Get a Supafone API key](https://labs.supafone.ai/console.html?mode=register) ## How it fits together 1. **Choose a speaking model.** It listens and talks to the caller. 2. **Enable Supervisor.** A separate reasoning model sends guidance during the call. 3. **Build in Agent Factory.** Keep the agent's job, tools and stages together for browser or phone calls. You select the model; Supafone routes calls to that saved selection. Use the [quickstart](quickstart.md) for a supervised agent or the [BYOK guide](https://labs.supafone.ai/docs/byok-providers/) for your own speaking and reasoning keys. ## Shared S2S interface Use the same `SupafoneS2S` interface for **Ultravox, OpenAI, Gemini, Grok, and Smallest AI Hydra**. Provider subclasses supply the selection while Agent Factory keeps the agent and phone identity. Ultravox stays the default; the native catalog offers six model choices. See [the common interface](unified-s2s.md) for the class contract and switching examples. Model changes apply to new calls, not a live-call handoff. Capabilities remain provider-specific: Hydra has no native transcripts and cannot change its persona or voice mid-session. Check credentials and test the selected provider before a customer call. ## Two voice-output choices in one Agent Factory **Ultravox + custom TTS** keeps the existing managed phone agent and lets a compatible Cartesia, ElevenLabs, Inworld, or Ultravox catalog voice supply its speech. **Native S2S** selects OpenAI, Gemini, Grok, or Hydra and uses that model's own voice list. Both are available through `SupafoneS2S`; external TTS is not a universal voice override for native models. [Compare both paths and create an Ultravox agent with custom TTS](voice-output-modes.md). ## Native realtime Agent Factory | Speaking provider | Supported model | Default voice | | --- | --- | --- | | OpenAI | `gpt-realtime-2.1` | `marin` | | OpenAI | `gpt-live-1` | `marin` | | Google | `gemini-3.1-flash-live-preview` | `Puck` | | xAI | `grok-voice-latest` | `eve` | | Smallest AI | `hydra-v1.0` | `sterling` | | Smallest AI | `hydra-v1.1` | `maya` | Each model has an adapter for authenticated browser previews and phone calls through **Supafone-managed phone, Twilio, Telnyx, Plivo, or SIP**. The catalog reports supported integrations; workspace credentials, model access, carrier setup, and a live test determine whether a particular deployment is ready. Start with your Supafone API key. For the selected model, Supafone uses its platform provider key when configured; you do not need to paste your own vendor key. An encrypted account BYOK key overrides the platform key for that provider. Check the runtime status before starting a call. A configured key alone does not prove provider permissions or call quality. ## What the harness keeps together | Layer | Your choice | Supafone's job | | --- | --- | --- | | Agent | Name, instructions, business knowledge, supported tools | Save one account-scoped agent and enforce tool authority | | Stages and team | Generated/custom stages, Manager and specialists | Validate legal transitions, saved facts, tool receipts and active permissions | | Speaking model | OpenAI, Google, xAI, or Smallest AI catalog selection | Translate audio, model events, and tool calls through its adapter | | Credentials | Managed platform key or optional BYOK | Keep secrets on the server and return masked readiness | | Delivery | Browser, managed phone, or a supported BYO carrier | Connect the same agent through the selected transport | | Supervisor | Managed or BYOK reasoning model | Coach all five speaking families; native delivery uses `check_guidance`, with Hydra context labeled model-reported | | Client | Python, TypeScript, REST, or dashboard | Use the same hosted agent API | A switch takes effect on a new session. Select a voice offered by the new model; voices, latency, and model behavior are provider-specific. The harness does not make every feature identical across providers. See [Shared runtime, Manager and teams](shared-agent-runtime.md) for executable configuration and capability limits. ## Choose the runtime for the job | Path | Use it for | Current boundary | | --- | --- | --- | | **Native S2S Agent Factory** | Choose the speaking model and share the workflow across browser and phone | Custom stages, Manager, coaching, widgets, opt-in recording and native handoff; carrier and provider limits apply | | **Managed compatibility Agent Factory** | Existing Ultravox-backed agents and compatible external TTS or live language/voice profiles | Used when `realtime` is omitted; managed or BYOK Ultravox | | **Supafone Supervisor** | Coach all five hosted speaking families or an existing supported stack | Enable and configure separately; native delivery uses tools, while standalone adapter capabilities vary | The native harness supports knowledge lookup, lead capture, scheduling, SMS/email, and configured custom tools through the server's allowed tools. [Read the full feature boundaries](realtime-agent-factory.md#feature-boundaries) before choosing the runtime for a customer workflow. Live language/voice routing and the 1,600+ TTS voice catalog belong to the managed compatibility path; native S2S uses each model's own voice list. ## Start building 1. [Quickstart](quickstart.md): create, preview, and switch an S2S agent. 2. [Agent Factory](agent-factory.md): understand the agent and runtime choices. 3. [Native S2S guide](realtime-agent-factory.md): model, credential, browser, and carrier contracts. 4. [Developer workflows](developer-workflows.md): repeat the workflow across your agents. 5. [Managed keys and BYOK](byok-providers.md): choose who supplies each credential. 6. [Framework coverage](framework-support.md): native model support versus Supervisor adapters. ## Cost comparison See [Pricing and Credits](pricing-and-credits.md) for current managed-runtime billing and credit behavior. BYOK usage is billed by the selected provider. Do not assume that every model, carrier, or native S2S feature has the same billing or inclusions as the managed compatibility stack.