nodecode-glasses

Smart glasses: a wearer's words and camera in, short replies out. A Nodecode rewrite of 100yenadmin/glasses-platform by 100yenadmin.

A pair of glasses carries a token. It says something, the agent answers on the lens in one or two short lines, it asks "what is this" and the camera takes a picture, it shares a picture it already took, and it receives what the agent pushes without being asked: a reminder, a scheduled job's words.

Get it

Tell the agent "add the glasses cell", or run nodecode add nc://glasses.

Set it up

Give each pair a token in channels.glasses — one token for a single wearer (owner:default), or devices for several, in the original's user:device=token shape:

{ "channels": { "glasses": {
  "devices": "ada:frames-1=first-token, ada:frames-2=second-token",
  "session": "glasses",
  "auto_capture": "keywords" } } }

A token is a key: whoever holds it has a chat with your agent.

Use it

The device talks to four verbs on the gateway's own port, under path_prefix and its token:

Verb	What it is
`GET /glasses/<token>/hello`	pairs the device and echoes the wearer it is
`POST /glasses/<token>/say`	`{"text": "..."}`, and `{"image": "<base64 jpeg>"}` when the glasses already took a picture; answers `{"id": ...}`
`GET /glasses/<token>/poll`	the frames the agent has said since the last poll: `response` for an answer, `push` for something unasked
`GET /glasses/<token>/photo`	a capture request the device has not answered, or `idle`
`POST /glasses/<token>/photo`	`{"request": "...", "data": "<base64>"}` answers it

A visual question ("what is this", "read this", "the one on the left") makes /say ask for a photo first while auto_capture is keywords; set it to off to capture only through the glasses_capture_photo tool, which saves the photo as a file the answer can name.

The gateway refuses any Host but loopback, so put an HTTPS reverse proxy in front of it (or nodecode link) before a device off this machine reaches it.

From the shell: /glasses shows the endpoint and the pairs, /glasses say TEXT sends one query down the same path a wearer's words take, and /glasses photo takes a picture from the last pair that spoke.