01
Offline wake word engine
Sub-10ms trigger detection tuned for always-on experiences without a network call.
Explore wake words ↗
MacroBot v2.4 Engine ReleasedZero Latency Edge Speech AI↗
Production-grade
VOICE INTELLIGENCE, WHERE THE DEVICE LIVES
Production-grade
voice AI for the edge.
Deploy localized wake word engine, offline intent recognition, and noise-robust speech recognition directly on ARM microcontrollers and edge gateways with zero cloud dependency.
✓Trusted by teams shipping at the edgeSOC 2 READY
// Initialize the voice runtime locally
auto engine = macrobot::Engine({
.model = "wakeword_v2_4.bin",
.device = "/dev/i2s0",
.latency = "realtime"
});
engine.on_intent([](Intent intent) {
auto payload = intent.to_json();
dispatch_local(payload);
});
engine.listen(); // 8.4ms p95 Listening on deviceRAM 12.8 MBCPU 4.1%
p95 latency8.4 ms
✓Cloud dependencyNone
LIVE SIGNAL01 / 04
BUILT FOR THE LAST MILEFrom first wake word to structured intent.
<10mstrigger detection
15MBRAM footprint
99.4%intent accuracy
01 / ENGINEERING SURFACE
See the intelligence
in motion.
A complete voice stack that runs close to the sensor — predictable, observable, and ready for production hardware.
SDK SHOWCASE / ARCHITECTURE PLAYGROUND
One runtime. Every edge.
Runtime healthy v2.4.0
RAM FOOTPRINT<15MBon Cortex-A53
P95 LATENCY<10mswake-to-intent
INTENT ACCURACY99.4%grammar-restricted NLU
DATA LEAVES DEVICE0%offline by default
EXECUTION PIPELINELOCAL / DETERMINISTIC / PRIVATE
01Audio inputmic / I2S / PCM
→02Feature extractionDSP + denoise
→03Local modelwake + intent
→04Intent JSONapp-ready output
02 / CAPABILITIES
Built for the
real world.
The hard parts of voice AI, solved for noisy rooms, constrained silicon, and teams who cannot afford a round trip to the cloud.
02
Speech-to-intent
Grammar-restricted local NLU converts voice into reliable, structured actions.
View intent schema ↗03
Edge DSP noise cancellation
Far-field voice enhancement that stays intelligible in the places your products live.
See DSP stack ↗04
Cross-platform SDK
Ship from C and Rust to Python, Android, iOS, Linux, and every major edge board.
Browse SDKs ↗05
Zero data leakage
100% offline processing for GDPR- and HIPAA-aligned product architectures.
Read our approach ↗06
Custom model training
Fine-tune models for your vocabulary, domain, accents, and product surface.
Train a model ↗03 / ENTERPRISE READINESS
Security is not
a feature flag.
MacroBot is designed for the environments where privacy, uptime, and control are non-negotiable. Keep your audio, models, and customer context on the device.
✓GDPRAligned by architecture
✓HIPAAOffline-ready processing
◈SOC 2Controls in progress
SUPPORTED HARDWARE5 / 5 ONLINE
◉Raspberry PiARM64 / Linux✓
▦STM32Cortex-M / bare metal✓
∞ESP32Xtensa / FreeRTOS✓
●Apple SiliconmacOS / iOS✓
x86x86_64Linux / Windows✓
“
MacroBot let us move the product from a cloud demo to a dependable appliance — without compromising the experience or our customers’ privacy.
2M+smart home & industrial units deployed
41%lower average power draw
0audio bytes sent to the cloud
READY WHEN YOUR HARDWARE IS
Put voice intelligence
where it belongs.
Bring your device, constraints, and roadmap. We’ll bring the runtime.
macrobot deploy --target edge
✓ runtime ready in 4.2s
// now listening locally
✓ runtime ready in 4.2s
// now listening locally