Voice with sub-second turns
Streaming speech-to-text, streaming responses, parallel tool calls, and barge-in so callers can interrupt. We build to an 800 ms turn budget and measure it on real calls, not a lab demo.
Voice and chat agents for the front line.
First-line support and sales over phone, SMS, and chat. Low-latency voice, your brand's tone, clear AI disclosure, and a clean handoff to a person whenever the caller wants one.
Channels and controls
Six parts of a front line callers trust.
One conversation stack across phone, SMS, and chat. Disclosure and handoff are part of the design, not settings.
Chat and SMS
The same agent, tools, and eval suite on your site, in your app, and over two-way SMS. Quiet hours and opt-outs handled per number.
WebIn-appSMS
AI disclosure on every channel
Every conversation opens by saying it is an AI assistant, in the caller's language. Chat shows a persistent label. SMS says it in the first message. We do not build agents that pretend to be people.
Your brand's tone
Your style guide written into the prompt: tone, vocabulary, what to soften, what to escalate. Your team signs it off before launch.
ToneVocabularyEscalation rules
Recording and privacy
Recording notices follow the caller's region. Card numbers go through keypad entry and are muted in recordings. Deletion requests run as a tracked workflow.
Regional noticesKeypad captureDeletion
Handoff to a person
One word routes the caller to your team. So can a topic, a sentiment drop, or a policy limit you set. The person picks up with the transcript, intent, and account already on screen.
The latency budget
Sub-second turns, measured on real calls.
Callers notice long pauses. We set a turn-latency budget before we build and measure it on recorded calls. No single fast model gets you there. A pipeline does — every stage streams into the next.
- Streaming speech-to-text, so the model starts before the caller finishes.
- Streaming responses, with the first sentence sent to speech as soon as it is ready.
- Parallel tool calls when an answer needs several lookups.
- Barge-in: the caller interrupts, the agent stops talking.
Honest by design
Disclosed as AI. One word from a person.
People should know when they are talking to software. Every conversation opens with a plain disclosure. Anyone who asks for a person gets one, with the context attached so they never repeat themselves.
- Disclosure in the first turn on every channel, in the caller's language.
- “Person”, “agent”, or 0 on the keypad routes to your team at any point.
- The handoff carries the transcript, intent, and account details.
- Disclosure wording checked with your counsel against local rules, including Article 50 of the EU AI Act.
800 ms
Turn-latency budget we build to, measured on real calls
First turn
AI disclosure on every channel
1 word
Routes the caller to a person, transcript attached
Every call
Transcript and outcome written to your CRM
Callers hear it's an AI in the first sentence, and anyone can ask for a person. Our team now takes the calls that need them.
Questions
What buyers ask us first.
Will callers know they're talking to an AI?
Yes. Every conversation opens with a plain disclosure, and we will not build one that hides it. Rules such as Article 50 of the EU AI Act require it.
What happens when the agent can't help?
It hands off to a person with the transcript, intent, and account details attached. You set the triggers: a request, a topic, a sentiment drop, or a policy limit.
Which phone and chat platforms do you work with?
Twilio or SIP trunks for voice and SMS, WebRTC for in-browser calls, and your existing help desk or chat widget where it has an API. We confirm fit for your stack in discovery.
How long until it takes real calls?
We plan for a pilot on a single line or queue in four to six weeks. Early on, overflow and edge cases route straight to your team.
How is it priced?
Each engagement is scoped and priced after a discovery call. Telephony, speech, and model usage bill to your own accounts.
Ready when you are
Tell us what your front line is drowning in.
A 30-minute discovery call. We'll tell you whether voice, chat, or neither should go first.