camera · diff

git:20260820.26f102d to git:20260904.6bef8df

29 added, 26 removed. Audit B to B.

---
name: camera
description: Camera control — snapshot, stream, and privacy toggle. Trigger on "what do you see", "look at this", "take a photo", "don't look", "stop looking", "stop watching", "stop staring", "camera off", "camera on", "give me privacy". MUST call [HW:/camera/disable:{}] or [HW:/camera/enable:{}] when toggling — never just reply with text.
---
# Camera
## Quick Start
Accesses the device's built-in camera at `http://127.0.0.1:5001` to take snapshots or check the environment. Only use when the user explicitly asks you to look at something.
## Already-captured frame (reuse, don't re-snapshot)
If the incoming turn contains a line like:
```
[vision-image] <absolute-path-to-a.jpg> (a photo was JUST captured ...)
```
a photo was **already taken** for this exact request by the realtime voice layer (it captured the frame, then handed the turn to you — e.g. it timed out mid-answer), and the OS layer delivers it **with this very message** — either as an `[image description]` line (when the main model is text-only, a vision model has already analyzed the photo for you) or as an attached image. **Answer the visual question from that description/attachment. Do NOT call `/camera/snapshot`** — re-snapshotting wastes time and may capture a different moment than what the user asked about. Do NOT read the path with a file tool — it is there for traceability only, and on text-only models a file-read image is silently dropped.
Only fall back to the snapshot endpoint below when there is no `[vision-image]` line.
## Capture Protocol
- Just call the snapshot endpoint — the server handles servo freeze, frame wait, and auto-enable if camera was disabled.
+ One call. It takes the photo, sizes it, and gives you back what is in it:
```bash
- curl -s "http://127.0.0.1:5001/camera/snapshot?save=true&width=768&quality=75"
+ curl -sX POST http://127.0.0.1:5000/api/vision/look \
+ -H 'Content-Type: application/json' \
+ -d '{"question":"<what the user asked>"}'
```
- Returns JSON: `{"path": ".../media/hal-snapshots/snap_1712567890123.jpg"}`.
- **Never hardcode a filename** — always read `path` from the response.
+ Returns `{"data":{"description":"...","path":"..."}}` — **answer from
+ `description`**. It is the only thing in this turn that actually saw the frame.
- `width=768&quality=75` shrinks the JPEG (~50–80 KB instead of ~300–500 KB at full 1920×1080) so vision LLM uploads + tokenizes faster. 768 px wide is still enough to read text on a laptop screen and recognize people/objects. Do NOT remove these unless you specifically need a larger image.
+ - No `description` field (the main model can read images itself) → open `path`
+ with your file/image tool.
+ - Any error → tell the user you couldn't see it this time. **Do not guess.**
- No need to aim servo or sleep before snapshot — the server freezes servos automatically for a stable frame.
+ The server handles servo freeze, frame wait, auto-enable if the camera was off,
+ and image sizing. Do not aim or sleep before calling it.
+ `GET http://127.0.0.1:5001/camera/snapshot` (HAL, port 5001) only writes a file
+ and returns its path — you cannot see that file. Use it only when you need the
+ raw frame for something other than looking at it.
+
## Never describe the view without an image
If you are about to say what you see, this turn MUST contain either a
- `[vision-image]` line or a `/camera/snapshot` call whose image you actually
- looked at. Describing the room from memory, from an earlier turn's photo, or
+ `[vision-image]` line or a `/api/vision/look` call whose answer you actually
+ looked at, or a `/api/vision/look` description. Describing the room from memory, from an earlier turn's photo, or
from a plausible guess ("same view — the desk, your screen…") is a fabrication,
even when the guess happens to be close. No image → say you'll take a look and
take one; never invent.
## Move first, then snapshot
When the request combines a movement and a visual question ("turn right, hold
it there, and tell me what you see"), fire the servo calls **with curl during
- the turn** (`POST /servo/aim`, `POST /servo/hold`), *then* snapshot. `[HW:...]`
+ the turn** (`POST /servo/aim`, `POST /servo/hold`), *then* look. `[HW:...]`
markers are executed only after your reply is composed, so a marker-based aim
would move the device *after* the photo — you would describe the old view.
## Workflow
- 1. Call `GET /camera/snapshot?save=true&width=768&quality=75` — **always call directly, never check /camera first**. The endpoint auto-enables camera if disabled.
- 2. Analyze the image and describe what you see.
- 3. Respond helpfully and specifically to the user's question.
+ 1. `POST http://127.0.0.1:5000/api/vision/look` with the user's question — **call it directly, never check /camera first**. It auto-enables the camera if disabled.
+ 2. Respond helpfully and specifically to the user's question, from the `description` it returns.
You also receive camera snapshots **automatically** as part of sensing events (`[sensing:*]` messages with images). You do not need the camera API for those — just look at the attached image.
## Examples
**Input:** "What do you see right now?"
- **Output:** `GET /camera/snapshot?save=true&width=768&quality=75` → analyze image. Say: "I can see your desk with a laptop and a coffee mug. Looks like a productive setup!"
+ **Output:** `POST /api/vision/look` → say: "I can see your desk with a laptop and a coffee mug. Looks like a productive setup!"
**Input:** "Is anyone in the room?"
- **Output:** `GET /camera/snapshot?save=true&width=768&quality=75` → analyze image. Say: "I can see one person sitting at the desk."
+ **Output:** `POST /api/vision/look` → say: "I can see one person sitting at the desk."
**Input:** "Take a photo" or "Send me a photo"
- **Output:** `GET /camera/snapshot?save=true&width=768&quality=75` → read `path` from JSON → describe what you see.
+ **Output:** `POST /api/vision/look` → say what `description` reports.
**Input:** (sensing event with image already attached)
**Output:** Do NOT call the camera API. Just look at the attached image and react.
## Tools
- **Bash** with `curl` for HTTP calls to `http://127.0.0.1:5001`.
+ **Bash** with `curl` — `http://127.0.0.1:5000` for `/api/vision/look`, `http://127.0.0.1:5001` for HAL camera control.
- ### Take a snapshot
+ ### Look at the scene
```bash
- curl -s "http://127.0.0.1:5001/camera/snapshot?save=true&width=768&quality=75"
- ```
-
- Returns JSON with the saved file path:
- ```json
- {"path": ".../media/hal-snapshots/snap_1712567890123.jpg"}
+ curl -sX POST http://127.0.0.1:5000/api/vision/look -H 'Content-Type: application/json' -d '{"question":"..."}'
```
- Without `?save=true`, returns raw JPEG bytes (used by web UI).
+ Returns `{"data":{"description":"...","path":"..."}}`. See *Capture Protocol*.
### Live stream
```bash
curl -s http://127.0.0.1:5001/camera/stream
```
Returns an MJPEG stream (`multipart/x-mixed-replace`). Only use when continuous video is needed. Prefer snapshot for one-time checks.
## Camera On/Off (Privacy Control)
Users can toggle the camera via voice or chat. Use HW markers — no curl needed.
### Disable camera
```
[HW:/camera/disable:{}]
```
The user wants privacy. Camera stays off until the user explicitly re-enables it (voice or web toggle).
### Enable camera
```
[HW:/camera/enable:{}]
```
### Trigger phrases (MANDATORY — must call HW marker, not just reply with text)
Any phrase meaning "stop looking" or "camera off" MUST trigger `[HW:/camera/disable:{}]`. Any phrase meaning "look at me" or "camera on" MUST trigger `[HW:/camera/enable:{}]`. Do NOT just acknowledge — you MUST include the HW marker.
| User says | Action |
|-----------|--------|
| "don't look" / "stop looking" / "stop watching" / "privacy mode" / "camera off" / "don't watch me" / "give me privacy" / "stop staring" | `[HW:/camera/disable:{}]` — MUST call |
| "look at me" / "camera on" / "you can look now" / "start watching" | `[HW:/camera/enable:{}]` — MUST call |
### "Look at ..." is ambiguous — route by what follows
The verb alone does NOT mean "turn the camera on". Only phrases about the *device's own
camera state* belong in the table above.
| User says | Meaning | Route to |
|-----------|---------|----------|
| "look at me" / "camera on" / "you can look now" | turn the camera back on | `[HW:/camera/enable:{}]` |
- | **"look at this"** / "look at what I'm holding" / "what is this" | a visual question about an object | **snapshot + analyze** (Workflow above) |
+ | **"look at this"** / "look at what I'm holding" / "what is this" | a visual question about an object | **`/api/vision/look`** (Workflow above) |
| "look at the desk / table / wall" | a fixed location | `servo-control` `/servo/aim` |
| "look at the cup and follow it" | a movable object to track | `servo-tracking` `/servo/track` |
**"Look at this" is a visual question, not a privacy toggle.** The user is holding something
up to be identified. Replying "Got it, camera on" answers a question they did not ask.
### Examples
**Input:** "Look at this" / "Look at what I'm holding"
- **Output:** `GET /camera/snapshot?save=true&width=768&quality=75` → analyze image. Say what the object is. Do NOT call `[HW:/camera/enable:{}]` — the snapshot endpoint auto-enables the camera.
+ **Output:** `POST /api/vision/look` → say what the object is. Do NOT call `[HW:/camera/enable:{}]` — the look endpoint auto-enables the camera.
**Input:** "Don't watch me"
**Output:** `[HW:/camera/disable:{}]` Got it, camera off. Just say "look at me" when you want me to see again.
**Input:** "Stop watching me"
**Output:** `[HW:/camera/disable:{}]` I'll look away. Let me know when you want me back.
**Input:** "Look at me"
**Output:** `[HW:/camera/enable:{}]` Camera back on!
### Auto-enable on snapshot (IMPORTANT)
**NEVER refuse a snapshot because camera is disabled.** The `/camera/snapshot` endpoint auto-enables the camera, captures the frame, then re-disables it automatically. Do NOT check `/camera` status before snapshot. Do NOT ask the user to enable camera first. Just call the endpoint.
## Error Handling
- If `/camera/snapshot` returns 503, tell the user: "The camera is not connected right now."
- If the API is unreachable, inform the user that the camera is temporarily unavailable.
- **Never check `/camera` status before snapshot** — just call `/camera/snapshot` directly.
- If a sensing event already included an image, do not call the camera API again.
## Rules
- **Just call `/camera/snapshot?save=true&width=768&quality=75`** — server handles servo freeze and camera enable automatically.
- **Always use `?save=true`** and read the `path` from the JSON response — never invent filenames.
- **Image delivery is handled automatically by the system** — do not manually send images via tools.
- **Never use the camera proactively without the user's request** — respect privacy.
- **Never disable/enable camera on your own** — only toggle when the user explicitly asks or when a system trigger requires it (guard mode, scene change).
- **Don't repeatedly snapshot without reason.**
- **Don't call the camera API when a sensing event already included an image.**
- **Prefer `/camera/snapshot`** over `/camera/stream` — simpler and sufficient for most tasks.
- When describing what you see, be specific and helpful.
- If camera is unavailable, inform the user clearly and move on.
## Output Template
```
[Camera] Action: {snapshot|stream|check}
Available: {yes|no}
Description: {what you see in the image}
```