recamera（recamera 音频）

reCamera Web API

Complete API reference for the reCamera (RV1126B) embedded camera platform. This skill enables agents to correctly construct HTTP requests, handle responses, and follow the interaction protocols required by the device.

Conventions

Base URL

All CGI endpoints are prefixed with:

/cgi-bin/entry.cgi/{api_category}/{resource}/{sub_resource}

Exceptions clearly noted per-endpoint (e.g. serial port uses /api/v1/..., file relay uses /storage/relay/...).

Authentication (CRITICAL — Token Must Be Captured and Reused)

Login flow (mandatory for all authenticated operations):

1. Fetch RSA public key: GET /system/key (no auth required)
2. Login: POST /system/login with {sUserName, sPassword} (password RSA-encrypted)
3. Extract token from response: On success (iStatus: 0, iAuth: 1), the HTTP response contains a Set-Cookie header like Set-Cookie: token=<jwt_value>; Path=/; .... You MUST capture the token value from this header.
4. Persist token for session: Store the extracted token value. All subsequent HTTP requests for this session must include the header: Cookie: token=<jwt_value>

For curl / shell scripts:

# Login and capture token from Set-Cookie header
RESPONSE=$(curl -s -D - -X POST http://{ip}/cgi-bin/entry.cgi/system/login \
  -H "Content-Type: application/json" \
  -d '{"sUserName":"admin","sPassword":"<RSA-encrypted>"}')
TOKEN=$(echo "$RESPONSE" | grep -oP 'Set-Cookie:.*token=\K[^;]+')

# ALL subsequent requests must carry the token cookie
curl -s -X GET http://{ip}/cgi-bin/entry.cgi/system/device-info \
  -H "Cookie: token=$TOKEN"

For browser automation (browser-use agent): After navigating to the login page and submitting credentials, the browser automatically stores the token cookie from the Set-Cookie header. However, you should verify the cookie exists by checking document.cookie for a token= entry after login succeeds. All subsequent same-origin XHR/fetch requests will include the cookie automatically (the app uses withCredentials: true).

IMPORTANT: If you receive an HTTP 401 response or a response body with "code": 401, the token has expired or is missing. You must re-login to obtain a fresh token.

Agent Playbooks (Mandatory Execution)

When user intent matches one of the following tasks, execute the corresponding workflow directly.

Playbook A: Discover reCamera Devices in a Subnet

Trigger intent examples: "scan subnet", "find recamera on 192.168.x.0/24", "discover devices in LAN"

Mandatory workflow:

1. Enumerate active IPs in the target subnet (e.g. ARP table / ping sweep / nmap host discovery).
2. For each active IP, send GET http://{ip}/cgi-bin/entry.cgi/system/key (no auth).
3. Classify as reCamera only if:

- HTTP status is 200, and - response contains field sPublicKey.

1. Return a structured result with:

- reachable_ips - recamera_ips - non_recamera_ips - unreachable_or_timeout_ips

Do not treat "TCP port open" alone as reCamera confirmation. /system/key verification is required.

Playbook B: Login and Persist Token for Follow-up Calls

Trigger intent examples: "login device", "call authenticated API", "configure after login"

Mandatory workflow:

1. GET /system/key to fetch RSA public key.
2. POST /system/login with {sUserName, sPassword} (RSA-encrypted password).
3. On successful login (iStatus: 0, iAuth: 1), parse Set-Cookie response header and extract token.
4. Persist token in session context.
5. For all subsequent HTTP requests in the same session, include:

- Cookie: token=<jwt_value>

1. If any request returns 401 / auth failure, re-login and refresh token, then retry once.

Do not continue authenticated calls without a captured token.

Configuration Update Pattern (Read-Before-Write)

CRITICAL: For any endpoint that supports both GET and POST/PUT:

1. First GET the current full configuration
2. Modify only the fields you need to change on the returned data
3. POST/PUT the complete modified configuration back

Never send a partial configuration constructed from scratch. The device expects the full config object and omitted fields may revert to defaults or cause unexpected behavior.

Field Naming

JSON keys use a type-prefix + camelCase convention:

Standard Response

Operation endpoints (POST/PUT/DELETE) return:

{
  "code": 0,
  "message": "success"
}

code: 0 = success. Non-zero = error (see error code ranges below).

Error Code Ranges

API Quick Reference

1. Authentication

Login response key fields:

• iStatus: 0=correct, -1=wrong password, -3=rate limited
• iAuth: 1=success, 0=fail, 2=must change password

2. Device Info

3. Live Video

stream_id: 0=main stream, 1=sub stream

4. Image (ISP) Settings

scene_id: 0=general, 1=day, 2=night

ISP Configuration Workflow:

1. GET /image/0 — fetch all params and determine current scene/profile in use
2. Select target scene profile (scene_id / iProfile: 0|1|2) based on current mode or user-specified mode
3. PUT /image/0/scene with {iProfile: 0|1|2} — enter that profile's live edit mode (5min timeout)
4. Adjust parameters (brightness/contrast/hue/saturation/sharpness etc.) via PUT /image/0/{scene_id}/{specific}
5. Send save/commit command: PUT /image/0/scene with {iProfile: -1} to exit edit mode and finalize changes

Mandatory sequence for image tuning tasks: "select mode -> enter mode edit state -> adjust params -> send save command". Do not skip the final save/commit step.

5. Recording

Trigger types: INFERENCE_SET, TIMER, GPIO, TTY, HTTP

Storage control actions: FORMAT, FREE_UP, EJECT, CONFIG, RELAY, RELAY_STATUS, UNRELAY, REMOVE_FILES_OR_DIRECTORIES

6. File Access (via Relay)

File access requires a relay session:

1. POST .../record/storage/control with sAction: "RELAY" — returns dRelayStatus.sRelayDirectory (UUID)
2. GET /storage/relay/{uuid}/ — list directories (Nginx autoindex JSON)
3. GET /storage/relay/{uuid}/{path} — download file
4. Relay auto-expires after 300s; re-request refreshes timeout
5. Video thumbnails: /path/to/.thumb/video.mp4.thumb.jpg (may not exist, implement fallback)

7. AI Model & Inference

Notification output modes: 0=off, 1=MQTT, 2=HTTP, 3=UART

8. WebSocket Endpoints

9. SenseCraft AI Cloud (ONNX to RKNN)

Base URL: https://sensecraft-train-api.seeed.cc (prod) / https://test-sensecraft-train-api.seeed.cc (test)

SenseCraft Auth Flow:

1. Redirect user to: https://sensecraft.seeed.cc/ai/authorize?client_id=seeed_recamera&response_type=token&scop=profile&redirec_url={your_url}
2. Receive token via callback
3. Backend resolves user_id: POST https://sensecraft-hmi-api.seeed.cc/api/v1/user/login_token with Authorization: {token} header

Conversion polling: 2s interval on /train_status; download only when status === "done"

Firmware Upgrade Flow

Two modes: local upload (resumable) and network download.

Network upgrade:

1. POST /system/firmware-upgrade?upload-type=network with {sReleaseURL} — returns version info + sConfirmToken
2. POST /system/firmware-upgrade?upload-type=network with {sConfirmToken} — confirm upgrade
3. GET /system/firmware-upgrade — poll download progress

Local upload (resumable):

1. POST /system/firmware-upgrade?upload-type=resumable with {iFileSize} — returns File-Id header
2. POST /system/firmware-upgrade?id={file_id} with binary chunks (Content-Range headers)
3. POST /system/firmware-upgrade?start={file_id}&md5sum={hash} — finalize upload

Factory Reset (Two-Phase Confirmation)

1. First POST /system/factory-reset — returns sConfirmToken (time-limited, ~1-5 min)
2. Second POST /system/factory-reset with {sConfirmToken} — executes reset

Key Constraints & Gotchas

• Read-before-write (GET-then-POST/PUT): When updating any configuration endpoint that has a corresponding GET method, ALWAYS fetch the current configuration first via GET, then modify only the needed fields on the fetched data, and POST/PUT the complete object back. Never construct a partial config payload from scratch — the device expects the full configuration structure and missing fields may be reset to defaults or cause errors
• JWT token expiration: The token obtained from login has a ~10-hour TTL. For long-running scripts or sessions, monitor for 401/auth errors and re-login to refresh the token
• RTSP capture with OpenCV: When using cv2.VideoCapture to grab frames from an RTSP stream, the first ~120 frames are stale buffered data. ALWAYS discard at least 120 frames before capturing a usable image (e.g. loop cap.read() 120 times, then take the next frame)
• BLC/HDR/HLC are mutually exclusive — only one can be "open" at a time in backlight settings
• Privacy masks: max 6 regions in OSD mask overlay
• ISP scene config: 5-minute inactivity timeout auto-exits live config mode
• Storage RELAY: requires slot state >= CONFIGURED; relay expires in 300s
• Model inference: model must have associated model info (JSON) before enabling
• InferenceSet + classification model: RegionFilter must be empty
• InferenceSet + detection model: at least one polygon required (default to full-frame [[0,0],[1,0],[1,1],[0,1]])
• Serial port endpoint does NOT go through CGI: /api/v1/device/serial-port
• File relay endpoint does NOT go through CGI: /storage/relay/...
• Device discovery: To detect whether a reCamera device is reachable at a given IP, send GET http://{ip}/cgi-bin/entry.cgi/system/key. This endpoint requires no authentication and returns immediately. A successful response (HTTP 200 with sPublicKey field) confirms the target is a reCamera device

Large Reference Reading Policy (Token-Efficient)

API_REFERENCE.md is intentionally large. Use it on demand instead of reading the whole file by default.

1. Default behavior: Do NOT read API_REFERENCE.md in full.
2. First-pass routing: Determine target module from user intent (auth/device/network/video/image/recording/storage/inference/websocket/cloud).
3. Section-first lookup: Read only the relevant section(s) for the current task.
4. Progressive expansion: If a field or constraint is still unclear, expand to adjacent subsection(s) only.
5. Full-file exception: Read the entire API_REFERENCE.md only when the user explicitly asks for a full audit/review/export, or when cross-module validation is strictly required.

Practical rule: prefer "minimum sufficient context" and keep reads scoped to the exact endpoint(s) being implemented or debugged.

Additional Resources

• For complete endpoint schemas with all fields and value ranges, see API_REFERENCE.md