VoiceRun Resources
Services VoiceRun Provides
Everything we offer today: what each service does, what’s included, and where the boundaries are. We keep this page current as the platform evolves.
Last updated August 31, 2026
The services
Voice Agent Platform
A managed platform for building and operating production voice AI agents. Agents are defined in code, so they are testable, auditable, and version-controlled, and they run on the world’s leading model providers.
What’s included
- Agent runtime and orchestration across speech-to-text, language model, and text-to-speech providers.
- Development tooling: CLI, SDK, templates, environments, releases, and deployment management.
- Testing infrastructure including regression tests, evals, and adversarial simulations.
- An improvement loop that uses production call data and human annotations to refine agent systems and fine-tune models. Models trained for a customer serve only that customer.
- Telephony integration: phone number provisioning, inbound and outbound calling, and bring-your-own-provider support.
- Session logs, transcripts, traces, and usage reporting for every call.
VoiceScore
Call quality measurement for voice AI. Every call a deployed agent takes is scored, rather than sampled, so teams can see how their voice systems actually perform in production.
What’s included
- Automated scoring of production calls against quality dimensions.
- Dashboards and reporting for tracking quality over time.
- Signals that feed the platform’s autonomous improvement loop.
Audio Control Plane
The connectivity and routing layer beneath the platform: telephony, SIP, and realtime audio transport, orchestrated across a roster of model and infrastructure providers.
What’s included
- Realtime audio transport between callers, telephony carriers, and model providers.
- Provider routing and failover across supported completion, speech-to-text, and text-to-speech services.
- Observability into latency, interruptions, and per-turn behavior.
VoiceRun Edge
The complete VoiceRun platform installed on your own infrastructure. One-click installation in AWS, GCP, Azure, or a private data center, using local models, external model services, or both.
What’s included
- Self-hosted deployment of the voice-agent platform in the customer’s cloud account or data center.
- Support for locally hosted models alongside external model providers.
- Suited to workloads with data residency, privacy, or regulatory requirements.
STT Router
Realtime speech-to-text routing built for voice agents. One key and one WebSocket provide access to Deepgram, OpenAI, ElevenLabs, Cartesia, Soniox, Google, and the VoiceRun STT Model.
What’s included
- A single realtime API in front of the supported speech-to-text providers.
- Unified authentication, billing, and usage reporting across providers.
- Provider and model selection per session, with a consistent message protocol.
VoiceRun STT Model
VoiceRun’s first-party speech-to-text model for voice agents, with per-turn context biasing, multilingual recognition, semantic turn-taking, and production guardrails.
What’s included
- Realtime transcription over the STT Router or the platform.
- Context biasing that can be updated mid-call for agent-specific vocabulary.
- Multilingual recognition and semantic end-of-turn detection.
Good to know
A few boundaries worth understanding before relying on the services in production:
AI-generated output is probabilistic
Transcription, language model responses, and synthesized speech are produced by machine-learning models and can contain errors, omissions, or unexpected behavior. Agents should be validated with the platform’s testing tools (regression tests, evals, and simulations) before and during production use, and customers remain responsible for how agent output is used in their business.
Third-party dependencies
The services orchestrate third-party model providers and telephony carriers. The availability, latency, pricing, and capabilities of those providers are outside VoiceRun’s control and can affect service behavior. Supported providers and models change over time; current rosters are listed in the documentation.
Not for emergency services
The services are not designed or intended for placing or receiving emergency calls (such as 911 or 112) and must not be relied upon for emergency communications.
Availability and support levels vary by plan
Uptime commitments, support response times, and usage limits depend on the customer’s plan or enterprise agreement. Specific service-level commitments for enterprise customers are defined in their contracts.
Language, region, and telephony coverage vary
Speech recognition and synthesis quality vary by language and audio conditions, and phone number availability varies by country and carrier. Regulated use cases may require configuration reviews with our team.
Staying informed
Technical documentation for every service lives at docs.voicerun.com. Material changes to the services are published on the Announcements page. The legal terms that govern use of the services are in our Terms of Service and Privacy Policy. For questions about scope, capabilities, or fit for a specific use case, contact us.