Inigami

Inigami bundles the components people normally stitch together by hand into a single C++ server: OpenCV computer vision (object detection, depth, segmentation, keypoints), FFmpeg video and audio processing, whisper.cpp speech-to-text, ImageMagick image transforms, and an NLP pipeline. They all sit behind a single TCP/JSON protocol on 127.0.0.1:10210, so you get one shape of request and one shape of response no matter which backend handles it.

Because the protocol is plain newline-delimited JSON, anything that can open a socket can drive it: a desktop Qt client, a CLI, a Python script, or an LLM through the included MCP tools layer (19 tools that proxy the same TCP protocol over stdio). The server binds loopback by default and ships with no auth, on purpose. It's meant to live next to whatever's calling it.

Vendored dependencies built from source via a Makefile + Ansible/systemd deploy. No cloud account required: spin it up on your laptop, on a homelab box, or in front of a model.