Frontier mode
Route any stage to leading cloud providers through PeachDesk-managed accounts. This is the fastest path to production quality, and it is always metered: usage draws down managed credits at published rates.
Platform
The voice AI platform you can own
PeachDesk is an AI voice agent platform for building, deploying, and operating voice agents across phone and web surfaces. Unlike metered orchestration platforms, it lets teams run any pipeline stage on managed frontier providers, their own provider keys, or a full open-model stack inside their own environment, mixed per stage per agent.
Own the stack. Do not just rent the pipeline.
Two model planes, one platform
Every voice agent is a pipeline: it listens, thinks, and speaks. Each stage of that pipeline needs a model, and you choose where each model runs. Frontier mode, Local mode, mix and match. One voice AI orchestration layer for your voice agent builder workflows, your AI calling platform operations, and everything in production.
Local mode
Run any stage on a full open-model stack inside your own environment, on GPU capacity you control. There is no per-minute provider charge for self-hosted inference; you pay for GPU capacity, and your call data stays in your environment for those stages.
Mix and match, per stage
For each agent, set the sourcing of speech-to-text, the language model, and text-to-speech independently: PeachDesk-managed providers, your own provider keys (BYOK), or self-hosted inference. One agent can reason on a frontier model and speak through your own stack.
Build, test, publish, operate
The platform covers the whole life of a calling agent, from the first draft to the ten-thousandth call.
Build
Design call flows on a visual canvas of 16 node types and conditional edges, start from 26 pre-built templates, or describe a use case and let the AI Agent Builder draft a starting agent. Explore the agent builder
Test
Test every agent four ways before a real caller hears it: text chat, AI-simulated callers, in-browser voice, and real phone calls.
Publish
Publish immutable agent versions. Every call is pinned to the exact version that served it, so what you tested is what callers get.
Operate
Connect 7 telephony providers or your own Asterisk PBX, score real calls with a configurable QA step, and track cost of goods, price, margin, and sourcing mode on every metered call. Explore telephony
Need production voice, not just a voice agent? PeachDesk Voice Studio clones voices, generates speech, and builds multi-voice timelines in the browser.
What happens on a call
A caller speaks. PeachDesk transcribes the audio, your workflow decides what happens next with the language model and your tools, the reply is synthesised to speech, and telephony or browser audio carries both sides of the conversation. Each of those stages can be sourced differently. Read the plain-English walkthrough of how PeachDesk works.
Know the cost before the call
Every agent shows an estimated cost per minute and latency before you run it, framed as an estimate, never a quote. After the call, every metered call records cost of goods, retail price, margin, and sourcing mode, so finance and engineering read the same numbers. See PeachDesk pricing for the full charging mechanism.
Hosted by us, or hosted by you
PeachDesk is available as a fully managed, hosted platform, or as a Docker Compose stack you run on your own infrastructure with billing disabled. Self-hosting is a product path, not an enterprise gate. Compare the two on the self-hosted voice AI deployment page, or control the whole platform programmatically through the developer platform: a versioned REST API, typed Python and TypeScript SDKs, and an MCP server for agent-driven authoring.
What owning the stack means
Own your cost
Choose the sourcing of every pipeline stage per agent and see the estimated cost per minute before a call runs.
Own your data
In Local mode, inference runs inside your environment, and call audio and transcripts need not leave it for those stages.
Own your margin
Operate the platform on your own infrastructure and see cost versus retail on every call.
What is a voice AI platform?
A voice AI platform is the software layer teams build production calling agents on. It combines orchestration (conversation flows, tools, knowledge bases), telephony and browser calling, testing, deployment, and cost controls in one system. PeachDesk is a voice AI platform that adds an ownable model plane: every pipeline stage can run on managed frontier providers, your own provider keys, or a full open-model stack in your own environment.
Can I mix cloud models and local models in one agent?
Yes. PeachDesk routes each pipeline stage independently: speech-to-text, language model, and text-to-speech can each use a managed frontier provider, your own provider keys, or self-hosted inference in your environment. A single agent can reason on a frontier model while its speech runs on a full open-model stack. Mixing is configured per stage, per agent.
Does PeachDesk work with my existing telephony provider?
PeachDesk connects to seven telephony providers: Twilio, Vonage, Plivo, Telnyx, Cloudonix, Vobiz, or your own Asterisk PBX. Phone numbers are bring-your-own: you register numbers with your provider, manage multiple telephony configurations per organisation, and route each inbound number to the agent assigned to it.
See the platform on your use case
Bring a call flow and we will walk it through the canvas, the sourcing modes, and the cost estimate.