FORERUNNER // FRONTIER AI RESEARCH LAB
We build agents that don’t start over every time you come back.
Forerunner is a frontier AI research lab. We work on what an agent perceives, what it remembers, how it speaks, and how it holds a long piece of work together, then ship that research as products: Mirari and HealthPass on iPhone, and Friday, our desktop agent, in early access.
01 · What we believe
Software forgets you the moment you close it. That’s the part worth fixing.
Every tool you use starts from zero. You re-explain the situation, re-attach the same file, answer the question you already answered last week. The machine has no idea who it’s talking to, so the work of remembering falls on you.
We build for the opposite. Something that already has the context, keeps it between sessions, and picks up mid-thought. You stop briefing it and start working with it.
02 · The work
Three problems we keep coming back to.
It takes in what’s in front of it
An agent that can see and hear its surroundings responds to what is actually happening, instead of waiting to be told.
It remembers what matters
Memory that outlasts the session, and knows the difference between a passing detail and something worth keeping.
It answers out loud
Speech that arrives while you’re still in the conversation, expressive, interruptible, and easy to talk over.
It picks up where you left off.
Mirari is the lab’s work in your pocket: agents that hold a personality, remember you between conversations, and answer out loud. On iPhone.
04 · Research
We publish the work.
Making an agent feel effortless is mostly research: keeping a model coherent across thousands of turns, deciding what is worth remembering, and answering out loud without a pause that breaks the conversation.
We write that up and put it online — papers with methods and figures, plus shorter engineering notes. If you want to check the reasoning, the material is there.
READ THE RESEARCH06 · Company
The facts.
FOUNDED
2024
HEADQUARTERS
San Francisco, CA
CONTACT
go@forerunner.ai
Build on the same stack.
The models behind our own apps reach developers through one API. Build on them, back the lab, or just say hello.