Supafone Labs#

Production infrastructure for voice agents that need to work after the demo.

Python SDK · TypeScript SDK · GitHub · Developer console · API reference

The so what#

A normal voice framework helps an agent speak. Supafone helps developers build, test, operate, supervise, and improve the entire voice product.

Use one Supafone API key for the production systems that otherwise become separate integrations: phone and WebRTC delivery, managed voices, grounded knowledge, verified tools, email, live supervision, recordings, transcripts, QA, and post-call evidence. Start with Supafone-managed infrastructure, or keep the realtime model, carrier, and speech providers you already use.

Follow the 10-minute quickstart · Create an API key · Open the workspace · Review framework coverage

What changes for the developer#

Voices without five separate integrations#

The managed catalog currently normalizes more than 1,600 voices across Cartesia Sonic, ElevenLabs, Inworld TTS-2, Deepgram Aura, and Ultravox's speech-native managed voices. Search and preview by provider, language, gender, and style, then deploy a compatible managed voice without adding that provider's API key. BYOK remains available for teams that need a specific provider account or private voice.

Cost comparison#

Supafone Cloud is $0.10 per connected minute for the standard managed call stack, including managed models, compatible voices, telephony, transcripts, Supafone Supervisor, and QA. The first five minutes are free; a 400-minute reload is $40. There is no seat fee.

That $0.10 pays for the production path, not only the SDK call that starts it. Supafone operates the carrier connection, compatible model and voice runtime, artifact pipeline, Supervisor inference, and QA record under one meter. It also removes the engineering and operational overhead of securing several vendor credentials, reconciling their usage records, and rebuilding the same control plane for each agent. Teams that already have better provider economics can use BYOK and keep Supafone as the common runtime and evidence layer.

Competitor headline rates often meter only one layer of the stack. This table keeps the published rate and the separately billed pieces visible:

At 10,000 connected minutes, Supafone's standard managed stack is $1,000. The same published-volume subtotal is $500 for Vapi before model, speech, and carrier usage; $750 for Deepgram before carrier; $800 for ElevenAgents before LLM and telephony; and $1,400 for Bland Start before carrier. The point is not that every workload has the same provider mix. It is that Supafone's headline rate describes the standard production stack, while many headline rates are a base layer that still needs several services and integrations.

Rates checked August 31, 2026. Vendor prices and inclusions change; follow the linked pricing pages before making a purchasing decision.

Why we built it#

A voice demo can be assembled quickly. A dependable voice product cannot. The production agent is split across a realtime model, telephony, TTS, STT, tools, retrieval, state, recordings, compliance, monitoring, and post-call workflows. Every vendor exposes a different event format, and the speaking model is still expected to notice and correct its own mistakes while talking.

Supafone Labs was built around the failures that appear at those boundaries:

The architecture#

Caller -> speaking agent -> tools and business systems
             |                         |
             +---- call events --------+
                          |
                          v
               canonical call state
                    |          |
                    v          v
              live supervision   call artifacts
                    |
          confidence + policy gate
                    |
              provider adapter
                    |
          silent bounded guidance
                    |
                    +------> speaking agent

The call never waits for supervision. If it is unavailable, late, or uncertain, the gate emits no directive and the original agent continues.

Two ways to use the package#

Supervise an agent you already run#

import supafone_labs

supervisor = supafone_labs.supercharge(my_agent)
result = await supervisor.observe(provider_event)

The package auto-detects supported agents when possible, normalizes their events, and returns the provider-appropriate action. Start with Supafone Supervisor, then check the framework coverage matrix.

Provision the complete agent#

import { Supafone } from "supafone-labs";

const supafone = new Supafone({ apiKey: process.env.SUPAFONE_TOKEN! });

const agent = await supafone.labs.agents.createInboundWithNumber({
  agentKey: "northline-intake",
  name: "Northline intake",
  description: "Understand the request and book the right next step.",
  number: { search: { areaCode: "415" } },
});

Agent Factory adds the plan, number, voice, stages, tools, call artifacts, and live supervision. Developers can inspect and edit the generated plan before creation.

Framework coverage#

The release gate covers fourteen runtime integrations, not a marketing-only logo list:

• Native control: Supafone Agent Factory, Ultravox, Vapi, OpenAI Realtime,

Grok Voice Agent, ElevenLabs Agents, Deepgram Voice Agent, and Inworld Realtime.

• Context owned by the developer: Retell custom LLM, LiveKit Agents, and

Pipecat. This is also the supported live-steering path when Gemini is hosted inside one of those frameworks.

• Observation or explicit hook: Gemini Developer Live, Bland, and Cartesia

Line.

• Extension path: GenericWebhookAdapter for proprietary systems.

Integration depth is different for each provider. The complete framework matrix shows the exact control, acceptance criterion, and managed-delivery status for every runtime.

Package surfaces#

Start here#

1. Read the production problems.
2. Follow the quickstart.
3. Understand Supafone Supervisor.
4. Review all supported frameworks.
5. Install the Python or TypeScript SDK.
6. Choose managed delivery or BYOK.
7. Run the voice-agent QA workflow.

Supafone Labs exists so developers can define the caller experience, tools, and safety policy while one framework handles the infrastructure around them.

View raw Markdown