EXCAR isn't a rendering or a pitch deck — it's running in a real 2015 Mazda 3 today, plugged into the OBD-II port, listening and talking back. Here's exactly what that looks like, and what's left to build.
The full pipeline — listening, thinking, and talking back — runs end-to-end on the prototype hardware right now.
Always-listening wake detection feeds a streaming speech-to-text → language model → speech pipeline, with overlapped playback so EXCAR starts talking before it's finished thinking.
Connected directly to the car's OBD-II bus — reading fuel level, speed, RPM, coolant temperature, and diagnostic trouble codes in real time, with EXCAR drawing power from the same port.
Conversations carry across drives. EXCAR restarts with a context-aware acknowledgment of how much time has passed — without ever announcing that it's "remembering" anything.
A companion app runs directly on the car's built-in screen (Mazda CMU), giving EXCAR a visual presence on the dashboard itself — not just a voice from a box.
A React Native app for setup, EXCAR's personality settings, and reviewing — or clearing — what EXCAR remembers. Currently in beta.
The voice pipeline, memory, and OBD-II integration layer are covered by an automated test suite — the kind of unglamorous groundwork that keeps a prototype from breaking every time we touch it.
Each stage builds on the last — hardware first, then the offline brain, then real drivers.
Core voice pipeline and OBD-II integration running on a Raspberry Pi 4 in a 2015 Mazda 3. Wake word, streaming conversation, persistent memory, and live vehicle data all working together.
Moving to more capable on-board hardware to support a fully local language model — so EXCAR keeps working in tunnels, parking garages, and dead zones, switching seamlessly between cloud and on-device.
A hands-on installation for ten Boston-area drivers. Each car gets a personal setup session, direct feedback channel, and a front-row seat to shaping what EXCAR becomes next.
Wider rollout across more vehicles and car models, with the plugin architecture extended beyond Mazda to validate "works in any car" across real-world hardware variation.
Opening EXCAR up to anyone who wants one — backed by a working product, a real install base, and a year of road testing behind it.
EXCAR's software is split into a car-agnostic core and small per-vehicle plugins, so supporting a new car model is an addition, not a rewrite.
Wake word detection, speech-to-text, the conversation engine, text-to-speech, and memory — identical in every car, regardless of make or model.
Per-vehicle integration — OBD-II reading, dashboard display, steering wheel controls — lives in a small plugin. Mazda is first; BMW and Toyota plugins follow the same pattern.
EXCAR runs on a cloud language model when connected, and falls back to a quantized on-device model with no perceptible change in personality when it isn't.
Join the waitlist for the Boston pilot, or follow along as EXCAR moves from one car to a fleet.