Book a call

Get access

An applied research lab building action AI

AI should do things for you, not just chat. We do original research on models that act, and we ship them as products: AGI-ACT, a small action model, AGI Agent, which works across any app, and AGI-RUN, an inference engine that runs it at the edge.

AGI, Inc. was founded in 2025 in San Francisco by Div Garg, Steve Frey and Naman Garg.

Talk to us

What we build

AGI-0

Our family of small action models, from 2B to 20B+ parameters. AGI-0 scores 100% on AndroidWorld, 116 of 116 tasks.

AGI Agent

The assistant built on AGI-0. It sees the screen, reasons, and acts across any app. No integrations, no APIs.

AGI-RUN

Our inference engine for phones, laptops, and edge hardware. Up to 9.9x faster than state-of-the-art engines, winning 58 of 60 benchmark cases.

AGI-0

Our family of small action models, from 2B to 20B+ parameters. AGI-0 scores 100% on AndroidWorld, 116 of 116 tasks.

AGI Agent

The assistant built on AGI-0. It sees the screen, reasons, and acts across any app. No integrations, no APIs.

AGI-RUN

Our inference engine for phones, laptops, and edge hardware. Up to 9.9x faster than state-of-the-art engines, winning 58 of 60 benchmark cases.

Values

Learn continuously

Learn continuously. The field changes every week. So do we.

Build trust

Build trust. With users, partners, and each other. Say it, then do it.

Pursue excellence

Pursue excellence. The bar is the best work in the field, not the last version.

Think first principles

Think first principles. Reason from what is true, not from how it is usually done.

Create delight

Create delight. Products people enjoy using, not just tolerate.

Move at the speed of light

Move at the speed of light. Ship fast, learn fast, fix fast.

Take ownership

Take ownership. See the problem, own the problem.

Learn continuously

Learn continuously. The field changes every week. So do we.

Build trust

Build trust. With users, partners, and each other. Say it, then do it.

Pursue excellence

Pursue excellence. The bar is the best work in the field, not the last version.

Think first principles

Think first principles. Reason from what is true, not from how it is usually done.

Create delight

Create delight. Products people enjoy using, not just tolerate.

Move at the speed of light

Move at the speed of light. Ship fast, learn fast, fix fast.

Take ownership

Take ownership. See the problem, own the problem.

Learn continuously

Learn continuously. The field changes every week. So do we.

Build trust

Build trust. With users, partners, and each other. Say it, then do it.

Pursue excellence

Pursue excellence. The bar is the best work in the field, not the last version.

Think first principles

Think first principles. Reason from what is true, not from how it is usually done.

Create delight

Create delight. Products people enjoy using, not just tolerate.

Move at the speed of light

Move at the speed of light. Ship fast, learn fast, fix fast.

Take ownership

Take ownership. See the problem, own the problem.

Team

Div Garg

CEO and Co-founder. Previously founded MultiOn, the first company to deploy computer-use agents in the browser. Adjunct CS professor at Stanford; created the first course on Transformers (10M+ views) and co-taught with Andrew Ng. 3K research citations and 10 patents in AI.

Steve Frey

Co-founder & President. Previously founded a fintech scaled to $500M AUM. Research at Stanford and Berkeley. Founding research at Eon, which collected the world's largest connectomics dataset, and at Kernel, a neurotech hardware lab.

Naman Garg

Co-founder and CTO. First engineering hire at MultiOn. Previously at Amazon.

Advisors and Investors

Advisors

Rich Miner

Founder of Android

Richard Socher

Recursive AI / AiX

Prof. Chris Manning

Stanford AI Lab

Prof. Stefano Ermon

Stanford AI Lab, Founder at Inception Labs

Investors

Menlo Ventures

Point72 Ventures

Angels

Charlie Songhurst

Meta board of directors

Linked In

Gautham Thambidorai

VP, DeepMind

Ali Rowghani

fmr. Managing Director at YC, COO at Twitter

Songyee Yoon

HP Board of directors

Eugenia Kuyda

Replika

Nikita Shamgunov

Neon

Vijay Krishnan

Turing

Partners

Qualcomm

Lenovo

Visa

FAQ

What does AGI Inc. build?

AGI, Inc. builds action AI: AGI-0, a family of small action models from 2B to 20B+ parameters; AGI Agent, an assistant that sees the screen and acts across any app with no integrations or APIs; and AGI-RUN, an inference engine for phones, laptops and edge hardware.

Where does AGI-0 run?

Both. AGI-0 runs on the device, on phones, laptops and edge hardware through AGI-RUN, and it can also run in a hybrid setup with cloud or on-prem components, fitted to your hardware and compliance needs. On-device keeps data on the device; hybrid adds capacity for heavier tasks.

How is user data handled?

Privacy is a design principle, not a feature. AGI-0 can be deployed on-prem or at the edge so data stays inside the environment you control. Our privacy policy describes what the AGI app collects and why.

Can AGI-0 be tuned to our devices?

Yes. AGI-0 can be fine-tuned to a device maker's specific app set and hardware. That is how we work with OEM and silicon partners such as Qualcomm.

How do teams integrate it?

There is nothing to integrate app by app. AGI Agent works from the screen, so it needs no APIs or per-app connectors. Device makers and platform teams work with us directly on deployment, and developer documentation lives at docs.agi.tech.

Who is behind AGI Inc.?

AGI, Inc. was founded in 2025 in San Francisco by Div Garg, Steve Frey and Naman Garg. Div previously founded MultiOn. The company is backed by Menlo Ventures, Point72 and angel investors from OpenAI, DeepMind and Meta.

What does AGI Inc. build?

Where does AGI-0 run?

How is user data handled?

Can AGI-0 be tuned to our devices?

Yes. AGI-0 can be fine-tuned to a device maker's specific app set and hardware. That is how we work with OEM and silicon partners such as Qualcomm.

How do teams integrate it?

Who is behind AGI Inc.?

What does AGI Inc. build?

Where does AGI-0 run?

How is user data handled?

Can AGI-0 be tuned to our devices?

Yes. AGI-0 can be fine-tuned to a device maker's specific app set and hardware. That is how we work with OEM and silicon partners such as Qualcomm.