open-source pan/tilt camera

A camera you can
tell where to look.

AI agents drive it over MCP. Home Assistant finds it on its own. It tracks faces, and it can learn your own objects, all on a XIAO ESP32-S3 Sense you print and build in an afternoon.

scroll ↓

01 · agents

Tell it where to look.

lynx runs an MCP server and a REST API on the device. Connect Claude or any MCP client, and it can aim the head, take a snapshot and say what it sees.

look at the desk and take a picture
look(pan: -40, tilt: 30)
snapshot(annotate: true)
A laptop, a mug and a plant. One person, seated on the left.

02 · home assistant

Found, not configured.

Install the integration from HACS and lynx turns up under Discovered. Paste its token and you get the camera, pan and tilt, presets, patrol, occupancy, and firmware updates from GitHub.

presetDesk
presetDoor
patrolOn
occupancyClear

03 · face tracking

It finds you. Then it follows.

Face detection (people or cats) runs on the camera itself while the stream keeps 20 fps. Follow mode keeps the nearest face centred, then goes back to rest when it's gone. No frame leaves the device.

~4per second

±90°pan

0cloud

04 · custom models

Teach it your own things.

Gather training pictures with lynx itself, train a model in Edge Impulse, and install the export with one command. It then detects your objects on the camera, just as it does faces.

just capture mugit scans the room, 30 shots
Edge Impulsetrain FOMO or a classifier
just model-add export.zipmodel: custom

05 · build it

Eight printed parts. One screw type.

Two MG90S servos, one USB-C cable for power and data, a microSD slot with a clear path in. Nine M2 × 8 thread-forming screws hold it together. About $43 in parts.

Build your own lynx.