Voice AI that sounds human,
without the spin
We build the voice agent platform we wished existed: fast enough for real conversations, honest about what it can do, and yours to control end-to-end.
Our mission
Make voice the easiest channel to automate well
Most "AI voice" still feels like a bot reading from a script. We think the bar should be a real conversation: low latency, natural turn-taking, useful tools, and a clean handoff to a human when it matters. Kalem exists to make that experience the default — for support teams, sales teams, clinics, dealerships, and the agencies that build for them.
What we believe
The principles that shape every product decision.
Latency is a feature
Anything over half a second feels off on a call. We optimise the entire path — telephony, model, tools — to keep response times near 320 ms so conversations actually flow.
Your keys, your data
BYOC is a default, not an upsell. Bring your own OpenAI, Twilio, and Stripe credentials. We don't train on your audio, transcripts, or knowledge base — ever.
Honest copy, real numbers
No fake testimonials, no inflated logos, no review counts we can't back up. If we say a number on this site, it's measured — or clearly labelled as a target.
Built for production
Multi-tenant from day one, scoped credentials, audit logs, retention controls, and webhooks that retry. The boring parts that decide whether voice AI survives a real customer rollout.
Whitelabel without compromise
Agencies and platforms can resell Kalem under their own brand, with their own billing credentials, their own pricing, and their own subaccounts — without us in the room.
Ship the obvious
A voice agent without WhatsApp, transfers, knowledge bases, and webhooks isn't really a voice agent — it's a demo. We ship the boring, expected pieces first, then the new ones.
What we ship
The product, in plain English.
Voice agents over phone & SIP
Inbound and outbound calling on Twilio numbers or your own SIP trunk. Real-time speech-to-speech via the OpenAI Realtime API — no transcribe-then-respond pipeline.
WhatsApp, with voice notes
The same agent answers WhatsApp text and voice messages, so you only configure the conversation once.
Knowledge bases & tools
Drop in PDFs, websites, or text. Agents search a per-account vector store. Webhook-backed tools let them book, transfer, look up, or update anything in your stack.
Multi-tenant & whitelabel
Run one platform for all your customers, or your customers' customers. Branding, billing, credentials, and limits scope cleanly per account.
Target voice response time, end-to-end.
Customer-owned credentials, by default.
Audio or transcripts used to train models.
Build with us
Start free with 100 minutes, or talk to us about whitelabel and enterprise.