hailo-ai

Build an AI agent application with LLM tool calling for Hailo-10H.

Build an LLM chat application for Hailo-10H.

Build a GStreamer pipeline application for Hailo accelerators.

Build a standalone HailoRT inference application.

Build a Vision-Language Model application for Hailo-10H.

Build a voice assistant with Whisper STT and Piper TTS for Hailo-10H.

Camera setup, USB/RPi configuration, and troubleshooting.

Detect and report events from video streams.

Add, modify, or remove AI analytics overlay (bboxes, masks, landmarks, etc.) on a hailo-media-library encoded stream. Two paths — (A) board-side burn-in via OverlayStage, (B) host-side drawing by analytic_viewer. Use when the user asks for any analytics overlay or "see the bboxes/masks/landmarks". MUST ask whether the overlay should be drawn on board (burned into UDP stream) or on host (display-only by analytic_viewer) before proceeding.

Add a new output stream (sinkN) to a hailo-media-library app — resolution, framerate, and encoder. Use when the user asks "add another stream", "add an FHD/4K/HD output", or "I need a Nth video output". Edits application_settings.json + the active profile JSON, generates per-sink encoder/osd/masking files, and pushes to the board.

Snapshot the H15 SBC's runtime health — chip temperature, power consumption, CPU load, DRAM use, NN core utilization, DSP utilization, and the running app's PID. Use when the user asks "is the board OK", "is it overheating", "how loaded is the chip", "why is FPS low", or before/after a long demo to check how hot it's running and how much it's drawing. Read-only over SSH; no app restart.

Establish a working SSH connection to the H15 SBC over ethernet. Use when the user asks to connect to the board or set up board access. Default is root@10.0.0.1, password `root`.

Cross-compile a hailo-media-library app for the H15 SBC using the Yocto SDK on the host. Use when the user has modified main.cpp (e.g. via /swap-model) and needs an aarch64 binary, or asks "build it for the board". Does NOT require building the whole library tree — links against the prebuilt libs in the SDK sysroot.

Push artifacts (binary, configs, HEFs) to the H15 SBC and verify the app runs. Use when the user says "deploy it", "push to the board", or "run it on the device". Coordinates backup → scp → permissions → smoke test.

Modify the pipeline of an existing hailo-media-library app — either by editing the medialib JSON on the board (via /add-stream when the change is "more streams" / "different resolution" / "different sink") or by editing `main.cpp`'s `create_pipeline()` and cross-compiling when the stage topology must change. Use when the user says "make this app do X", "add a stage to X", "join Y into the pipeline", "make detection app also do Z", "stream out the AI metadata to another sink", or "run N streams instead of 1". Triages JSON vs C++ first; only goes to /cross-compile + /deploy on the C++ path.

Explain how a hailo-media-library application's pipeline is structured — sources, AI stages, sinks, and the medialib config that feeds it. Use when the user asks "what does this app do", "how is X pipeline wired", or before they modify an app. Reads the app's main.cpp, traces generate_*_pipeline calls, and maps it to the active medialib config.