Catch's AI assistant now makes real phone calls on your behalf -- and says it's already booked 12,000 meetings doing it

Catch, a Tel Aviv startup, emerged from stealth on September 3 with a $5 million seed round -- co-led by Entree Capital and Pitango, with Seedcamp and Factorial Capital also participating -- and a product built to do something most AI assistants still won't: place a real phone call on your behalf and talk its way through an automated menu to reach an actual person. The company, founded by CEO Nir Sabato and CTO Yoav Ramon, says it has already run quietly for four months with hundreds of paying customers before today's public launch.

The pitch is a single assistant that runs a calendar, triages an inbox, books travel, and makes outbound calls -- to restaurants, clinics, hotels, and airlines -- connected across Slack, WhatsApp, iMessage, email, and phone, working inside the accounts a user already has rather than a new dashboard to check. Catch says it can talk its way through an automated phone tree to reach a human, identifies itself as software calling on someone's behalf, and dials only on instruction; it returns a call summary and full transcript afterward. Pricing is a flat $99 a month, voice included, no per-call fees, with a 10-day free trial.

The security architecture is the part aimed at the objection every version of this pitch eventually meets: why would anyone connect an AI system to their email, calendar, and phone line all at once? Catch says it is SOC 2 compliant and independently audited, and its answer beyond the certificate is what it calls "splinter agents" -- rather than one general-purpose assistant with standing access to everything, each task spins up its own narrow agent scoped to only the data and permissions that task needs, so a restaurant-booking call can't also read three months of email. Whether that architecture holds up under real adversarial testing is not something a company's own launch materials can settle -- it's a claim about internal design that an outside security audit, not a product page, would need to confirm.

“Catch works alongside business leaders the way a great human assistant would, but 24/7, with no dip in attention and fully secure.” — Nir Sabato, Catch CEO and co-founder

On traction, Catch's own numbers don't fully agree with each other depending on where you read them this week. The company's press materials, cited across this week's coverage, put the total at 12,000-plus meetings scheduled, 20,000-plus tasks delegated, and 200,000-plus emails handled. Catch's own live site, checked the same day, shows a lower meetings figure -- 5,000-plus -- while matching the other two numbers exactly.

Catch isn't entering an empty market. AI scheduling and inbox assistants are a crowded category, and Catch's own comparison page names ten rivals -- among them Lindy, Motion, Fyxer AI, Reclaim, and Poke -- and draws the line at exactly the feature this launch leads with: voice. By Catch's own account, most of that field stops at drafting an email or proposing meeting times, a text-generation problem large language models already handle well. Dialing a real business, sitting through hold music, and improvising through a receptionist's actual questions is a harder, higher-stakes capability most competitors have avoided shipping as a default feature rather than a narrow demo -- though it's worth being clear that the comparison itself comes from Catch's own marketing, not an independent test of the other nine tools.

The harder question isn't the exact count -- it's who else is on the other end of an automated call that only one party agreed to. Catch says the assistant identifies itself as software the moment a human answers, which addresses transparency in the moment. It doesn't address consent before the moment: a restaurant host, a clinic receptionist, or an airline agent didn't sign up to have their call handled by an AI system, the way the Catch subscriber did. (This is the same tension every AI-calling product runs into -- the paying customer opts in, but everyone that customer's assistant calls on their behalf did not.)

None of this is unique to Catch -- AI scheduling and inbox tools have existed for years, and at least one rival startup let its own AI agent run its $100 million fundraise this summer as a publicity move. What's new here is narrower and more concrete: a real outbound phone call, not just a calendar invite, at a flat consumer-style price, backed by a security model the company is willing to name in public even if it hasn't yet been independently tested. Four months of quiet paying customers before today's launch is a genuine signal the product works well enough for someone to keep paying for it. It isn't yet a signal that the security architecture, or the anonymization behind the training-data pipeline, will hold up the way the launch materials say it will -- those are claims a $5 million seed round buys the company time to prove, not evidence that it already has.