Lisn · by Ground State Foundry

An AI that talks to your customers, wherever they are.

Lisn talks and chats with visitors on your website, answers from your own business knowledge, captures the lead, books the appointment, and hands over to a person when it should, in English and Arabic. Phone lines are next. It gets better by talking to other agents, before it talks to your customers.

The most stable state your business can run in.

How Lisn goes live

Five steps from connecting your channels to an agent that keeps improving on its own.

  1. 01ConnectPaste groundstatefoundry.com/skill into your coding agent (Claude Code, Cursor and others) and it installs Lisn in minutes. Or we set it up for you.
  2. 02TeachLisn learns your services, prices, hours, and FAQs from your own website. Add the answers your best people give.
  3. 03PractiseLisn talks to simulated customers first, and every conversation that goes wrong becomes a fix.
  4. 04Go liveIt starts answering real visitors and callers, and hands over to a person when it should.
  5. 05ImproveIt keeps practising against new situations, so it gets better every week.

What Lisn is built on

How Lisn works →
Lisn: phoneIn production

Production voice-agent platform

A multi-tenant platform that runs AI phone agents over real telephony, on the phone numbers a business already has.

  • Provider-agnostic: works across Twilio, Plivo, and Exotel
  • Campaign orchestration and a visual flow builder
  • Built so callers are not left in silence while the agent thinks

A voice agent that demos your product

An embeddable voice agent that walks a visitor through a live, interactive product demo on any website.

Try Lisn live

Agents that call agents

A simulation suite where one AI agent calls another through the real phone stack, so a voice system is tested before real customers reach it.

How it works →

What we make

Lisn

Early access

An AI that talks to your customers, wherever they are.

Bubbles

Open source · MIT

A terminal tool that runs a whole fleet of AI coding agents.

the lab behind it

Our system rediscovered a classic 48-multiplication method on its own. Then it went lower.

We pointed Darweel, our evolutionary framework, at 4x4 matrix multiplication and counted the multiplications each candidate used. Programs competed, mutated, and reproduced. Without being shown the answer, it arrived at Winograd's 48. A descendant then learned to skip multiplications by zero, which brought the average below 48 on the test set. The same lab builds the simulations Lisn practises against before it talks to your customers.

Strassen, 4964naive56early49Strassen48Winograd47.94zero-skip
Best candidate at each stage of our runs, in multiplications per 4x4 product.
64
schoolbook method
49
Strassen, applied twice
48
rediscovered by our system, matching Winograd
47.94
average across 500 test cases, all correct

Scalar multiplications per 4x4 product. 47.94 is a measured average over 500 test cases; the worst case is still 48. Winograd's 48 relies on the numbers commuting, so it applies to ordinary numbers, not to every kind of matrix algebra.

Who is building Lisn

Rishi Raj

Rishi Raj

Founder & Chief Scientist

Applied AI and research. Built voice agents that run in production over real telephony, an open-source system for running fleets of AI agents, and a framework where code evolves its own algorithms. Leads Lisn's research and product.

LinkedIn ↗
Devashish

Devashish

AI Advisor & Strategist

Advises on AI strategy and solution architecture end to end, from the model and the data pipeline to the product a customer actually uses. Keeps the hard technical choices honest.

LinkedIn ↗

Put Lisn where your customers already are.

We are onboarding a small group of early teams. Tell us where your customers reach you, and we will set Lisn up there first.

Ground State Foundry · Meydan Free Zone, Dubai