Books into Google Calendar and Microsoft 365 · uses your tools over MCP · runs in CH, EU and US regions — see where your data goes
How it works
Every turn runs through transcription, a semantic turn detector and a fast talker model. Anything slow — a calendar lookup, a CRM write, an MCP tool — runs in the background while the conversation keeps going.
An inbound call on your number, an outbound call from a campaign or the API, or a test call from the browser.
Speech is transcribed as it arrives, and a semantic turn detector decides when the caller has actually finished — not a fixed silence timer.
A latency-optimised model owns the conversation. It answers directly, or starts a tool job and keeps talking.
Questions, clarifications and small talk are answered from the agent’s own knowledge — no tool round-trip in the way.
Calendar, CRM or any MCP tool runs asynchronously while the talker says something useful — never dead air, never the same filler twice.
The reply is synthesised sentence by sentence and starts playing before it is finished — the result lands in the call record.
Latency budget per turn (design target)
Stages overlap — speculative generation starts before the caller has finished — so the total is less than the sum. Every call records its own per-stage timings.
Why this one
Self-serve voice platforms are fast to try and hard to deploy: US-hosted, tools bolted on through webhooks, compliance left to your prompt. Enterprise suites solve that for six figures a year. This is the missing middle — resident, MCP-native, compliant, and priced per minute.
Most self-serve voice platforms run every call through US infrastructure. Here a Swiss call is processed in Switzerland and an EU call in the EU — transcription, model and voice alike.
How · All inference routes through a residency-aware gateway: region-pinned hosts per jurisdiction, no cross-border fail-over, and a PHI-grade path where you need it.
An agent that can only talk is a voicemail with better manners. Give it your calendar, CRM and ticketing — the same way you give tools to any AI agent.
How · Register any MCP server once; choose per agent which tools it may use, which need a spoken confirmation, and which run in the background. Tool results are screened for prompt injection before the model sees them.
Phone rules differ by country and the fines are real. The guardrails should not depend on everyone remembering them in a prompt.
How · The runtime speaks the AI disclosure first, checks consent, do-not-call lists and local calling hours before dialling, and writes an opt-out to suppression the same turn.
Roles
Every role runs on the same runtime — the same voice pipeline, tools, calendar and compliance gates. What changes is the goal, the script and the tools you allow.
Answers every call, books, routes and takes messages — 24/7.
Learn moreCalls warm leads within minutes, qualifies and books the meeting.
Learn moreBooks, confirms and rebooks appointments — and cuts no-shows.
Learn moreInvites, confirms RSVPs and fills your events.
Learn moreChecks in after a sale or visit and captures structured answers.
Learn moreUnderstands the problem, opens the ticket, routes the urgent ones.
Learn moreWhat it will not do
It says it is an AI in the first sentence of every call. It does not dial people who opted out, numbers outside local calling hours, or — in Germany, France and Switzerland — anyone without a recorded consent. Robocalls and spoofed caller IDs are not a use case; they breach the terms.
Signals
Each call stores its transcript, the actions the agent took, the consent basis it dialled on, its cost and its latency per turn — so any question about a call has an answer.
Runs on
Runs on this infrastructure — not a partnership claim.
For developers
A versioned REST API under /api/v1/* to create agents and place calls, result callbacks when a call ends, and an MCP server (place_call, get_call, book_appointment and more) so any AI agent can make and read phone calls — or skip the code and use the dashboard.
What it will and won't do
The rules aren't a policy page nobody reads — they're enforced in the same path every call takes.
Every call opens by saying it is an AI assistant for your company (EU AI Act Art. 50). The runtime prepends it — a prompt cannot remove it.
Each outbound call passes a per-country rule table: consent on record, do-not-call lists, and the callee’s local calling hours. Outside the window, it is scheduled, not dialled.
MCP server secrets and calendar tokens are encrypted at rest, decrypted only for the call that needs them, and never sent to a browser.
Consent, opt-outs and do-not-call hits are append-only evidence per contact — source, time and basis — so you can show why a call was placed.
It will
It won't
FAQ
The short, honest answers on what it does, where it runs and what it costs.
Three ways in, pick whichever fits:
One all-in price per connected minute — from $0.19 on the free plan down to $0.13 at scale — including transcription, the model, the voice and standard termination. Numbers are $3/month (local) or $6 (mobile, toll-free); unanswered calls are free.
Still have a question? Read the docs or see every answer.
Free to start. Describe your company, pick a role, and talk to your agent in the browser in minutes — then give it a number, your calendar and your tools.
In-region processing · AI disclosure on every call · Per-minute, all-in