About
We're building the IDE that listens.
eri started as a side project: one engineer, a mic, and a hunch that staring at a blinking cursor was no longer the fastest way to write software. Three drafts later, it turned into something we wanted to ship.
Why eri exists
The first wave of AI editors made autocomplete dramatically better. The second wave gave us chat sidebars. Both still assume the human is the one driving the keyboard. We think that's the wrong frame. The keyboard is a 19th-century interface. Voice plus gaze plus a model that understands the whole repo is something else entirely — a pair programmer who is awake, attentive, and never asks you to rephrase.
eri is the first IDE built around that idea from the inside out. The runtime, the editor, the agent loop, the gaze tracker, and the voice pipeline are all our own code. There is no fork. There is no extension wrapping someone else's editor.
What we believe
Voice is the input layer
Typing is a bottleneck most of the day. Talking matches the speed of thinking. eri turns the half-formed sentence in your head into a working diff before you'd have finished the function signature.
Diffusion is the model
Autoregressive models write code one token at a time. Diffusion models write the whole file at once and then sharpen it. For edits across many lines, this is the difference between waiting and shipping.
Your machine is the runtime
Voice, gaze, and the editor itself stay local. Only the model calls leave the device, and you choose where they go. We sign the eri Engine calls; everything else can be bring-your-own-keys.
Who we are
A small team building in public. We post the weeks that went well and the weeks that didn't. We answer email. We ship on Fridays anyway. If that sounds like a place you'd want to work, the careers page is one click away.
Headquartered in San Francisco. Distributed across four time zones. Hiring in two more.