| name | rhombus-deployment-context |
| description | Generate, query, and analyze Rhombus deployment context — a cached snapshot of every location, camera, and still image across an organization — and use that context to review footage. Covers the `rhombus context`, `rhombus analyze`, and `rhombus stitch` commands. Use when the user asks about deployment context, camera snapshots, "what are my cameras seeing", wants to generate or refresh a site overview (`index.md`), asks Claude to analyze alert footage or a time window on a camera, wants to stitch multi-camera clips into a single review video, mentions incident review, wants to "pull video from last night", or asks to understand motion/activity in a specific time range. Also trigger for `--lan` mode questions and anything involving `~/.rhombus/context/`.
|
| argument-hint | [context|analyze|stitch] [args] |
| allowed-tools | Read, Grep, Glob, Bash |
Rhombus Deployment Context + Footage Analysis
Three CLI commands that work together to give Claude a working knowledge of a Rhombus deployment and let it reason over footage.
rhombus context — build a snapshot
Generates a local cache of locations, cameras, hardware, and per-camera stills. Claude reads the cache to understand what each camera covers without re-calling the API.
rhombus context generate
rhombus context generate --lan
rhombus context generate --concurrency 8
rhombus context location "Main Office"
rhombus context camera "Front Door"
Output files (in ~/.rhombus/context/<profile>/):
index.md — Compact deployment reference. Read this first when the user's question needs site-wide awareness.
manifest.json — Full machine-readable manifest.
stills/<camera>.jpeg — One still per camera.
Usage pattern: When the user asks about their deployment, Read ~/.rhombus/context/<profile>/index.md. When they ask about a specific camera, Read the matching still and any notes in the index.
rhombus analyze — describe what happened
Extracts and analyzes frames from alert clips or camera footage over a time window.
rhombus analyze alert "ALERT_UUID"
rhombus analyze footage "front lobby" --period "yesterday between 8am and 9am"
rhombus analyze footage --location "Main Office" --start 1700000000000 --end 1700003600000
rhombus analyze footage --location "Office" --period "last hour" --lan
rhombus analyze footage "parking lot" --period "today" --include-motion
rhombus analyze footage "parking lot" --period "today between 6am and 7am" --fill
rhombus analyze alert "ALERT_UUID" --raw --output /tmp/frames
rhombus analyze footage "lobby" --period "last hour" --raw
When to use: User asks "what happened…", "show me the activity at…", "describe the alert…". Prefer this over raw clip download when the user wants an answer, not the file.
rhombus stitch — chronological multi-camera review
Downloads clips for detected events and stitches them into a single chronological video. Concurrent events from multiple cameras are shown in a grid with timestamp overlays. Uses LAN streaming when available.
rhombus stitch --camera "front lobby,parking lot" --period "yesterday between 6am and 7am"
rhombus stitch --location "Main Office" --period "last night between 10pm and 6am"
rhombus stitch --camera "entrance" --start 1700000000000 --end 1700003600000 --buffer 10
rhombus stitch --location "Office" --period "today" --include-motion
rhombus stitch --location "Warehouse" --period "today between 8am and noon" --output incident-review.mp4
When to use: User wants a single video they can watch later — incident review, shift-handover summaries, insurance submissions. Prefer analyze when they just want an answer.
Typical flow for an incident review
rhombus context generate (if stale) so Claude knows the deployment.
rhombus analyze alert <UUID> — first pass for the triggering event.
- If the answer needs wider context:
rhombus analyze footage --location <site> --period "around <time>" for co-occurring activity.
- For a shareable artifact:
rhombus stitch --location <site> --period "<window>" --output review.mp4.
Slash commands
This plugin ships three slash commands that wrap the above:
/rhombus-analyze <alert-uuid | camera time-range> — fires analyze.
/rhombus-context-refresh — regenerates the full context (slash-only; writes to disk).
/rhombus-watch <camera> — opens live footage in the browser (slash-only; browser side effect).
Tips
- First context generation is slow (one still per camera, downloaded over the internet). Use
--lan if you're on the same network as the cameras.
- Keep the context fresh. Regenerate after major deployment changes (new cameras, reassigned locations). The
index.md footer includes the generation timestamp.
--raw mode is for when you want to pipe frames to a separate tool (your own model, a labeling flow) rather than have Claude describe them.