Will this model fit on this chip? Answers it before you flash. 31 boards, byte-exact memory from the real TFLite Micro runtime, native and in the browser. Quantization previews and a CI action.

What inference actually costs on real silicon. The ground truth mcufit is checked against. 54 measurements across 4 models and 3 boards, per model and per layer, timed on hardware and not estimated. Anyone can add a board.

Format-aware compression, without the expertise. Point it at your data. It writes the OpenZL description, trains a compressor, verifies the round trip byte for byte, and reports the savings against zstd and gzip.

Speech recognition and tool-calling models on embedded Linux boards, under 200 MB of RAM, no cloud round trip.
Mostly the memory arithmetic: what fits, what an ARMv7 core can hold in real time, and when the honest answer is that it does not.

nlohmann/json		JSON for Modern C++
yhirose/cpp-httplib		A C++ header-only HTTP/HTTPS server and client library
livekit/agents		A framework for building realtime voice AI agents
cactus-compute/needle		14MB foundation model for phones, wearables, smart home and robots
growthbook/growthbook		Open source feature flags, experimentation and product analytics
Serial-Studio/Serial-Studio		Telemetry dashboard for UART, BLE, MQTT, Modbus and CAN bus
Openpanel-dev/openpanel		Open-source web and product analytics, an alternative to Mixpanel
0xShug0/audio.cpp		Pure C++ inference engine for audio models, powered by ggml
dearlordylord/huly-mcp		MCP server and CLI for the Huly platform

loan-emi-negotiator	an agent that drafts a lender-specific rate-negotiation email
docpilot	document collection loop for vehicle-loan officers
fireAlertTelegram	ESP32 fire alert that pings you on Telegram
LaserTurret	laser module on a servo gimbal, aimed with a PS2 joystick