Voice AI glossary
Ten terms cover most of what buyers ask about voice AI. Each definition stands alone, and each ends with how the term shows up in PeachDesk. Part of the Resources library.
BYOK
BYOK (bring your own keys) means you supply your own API keys for speech and model providers instead of using the platform's managed providers. You pay those providers directly at your negotiated rates, and your usage stays out of the platform's billing path. In PeachDesk, calls on your own keys never consume PeachDesk credits and are never blocked by PeachDesk billing.
The voice AI BYOK page covers setup and sourcing details.
Frontier mode
Frontier mode is PeachDesk's operating mode that routes a pipeline stage to leading cloud model providers, metered per use. It gives an agent the strongest general-purpose models with zero infrastructure to operate, and the meter runs per call. Frontier mode is one of two model planes in PeachDesk; the other is Local mode.
The PeachDesk pricing page shows how metered usage is billed.
Local mode
Local mode is PeachDesk's operating mode that serves a pipeline stage from a full open-model stack running on GPUs in the customer's own environment. There is no per-minute provider charge for locally served inference; the customer pays for GPU capacity. Audio and text for those stages stay inside the customer environment.
The Local mode page covers deployment and trade-offs.
Mix and match
Mix and match is PeachDesk's per-stage, per-agent provider selection: speech-to-text, language model, and text-to-speech each independently run on PeachDesk-managed providers, your own keys, or your own infrastructure. A single agent can use a frontier model for reasoning and leading open models for speech, decided stage by stage.
The How PeachDesk works walkthrough shows per-stage routing.
Orchestration
Orchestration is the layer that coordinates a voice agent's models, call flow, tools, and call control: what the agent says, when it looks up knowledge, when it transfers to a human, and how the call ends. In PeachDesk, orchestration is the visual workflow, a canvas of typed nodes and conditional edges that defines the conversation.
The voice agent architecture explainer shows where this layer sits.
PeachDesk-hosted
PeachDesk-hosted is the fully managed cloud deployment of PeachDesk: the platform, dashboard, and managed model providers run for you, and usage is metered. It is the fastest way to start; teams that need infrastructure control choose the self-hosted deployment of the same platform instead.
The deployment options page compares the two ways to run PeachDesk.
Self-hosted
Self-hosted means running the full PeachDesk platform on your own infrastructure as Docker services: the API, dashboard, database, and object storage under your control, with Local mode inference on your own GPUs as an option. Self-hosters still pay their own providers and infrastructure; no deployment makes every call free.
The self-hosted voice AI page covers what runs where.
Sourcing mode
A sourcing mode is PeachDesk's per-request classification of who provides and who bills a pipeline stage: PeachDesk-managed providers, your own keys (BYOK), or your own self-hosted inference. Sourcing modes are set per stage, per agent, and they decide both the cost of a call and where its data goes.
The voice AI cost per minute explainer shows how sourcing changes the math.
Voice agent
A voice agent is software that holds a spoken conversation with a person over a phone call or browser audio. It listens with speech-to-text, decides what to say with a language model, speaks with text-to-speech, and connects through telephony or WebRTC. PeachDesk is a platform for building, testing, and operating voice agents.
The voice AI platform overview covers the building surface.
Voice Studio
PeachDesk Voice Studio is the hosted suite for creating voice assets in the browser: generate speech, clone and manage voice profiles, apply effects as versioned takes, and build multi-voice timelines. Voice Studio voices plug into the same platform that builds and deploys calling agents, and assistants can drive it through its own MCP server.
The Voice Studio page tours the suite.