Argus Intelligence Dashboard

Project Context

The repo is at ~/argus. Never search in /home/volta root — the project is always here.
Main app code lives in argus-app/src/.
Always confirm repo path before git operations.

You are already on the production droplet. No SSH. Run docker commands directly.
The backend IS the Docker stack on this droplet, fronted by Cloudflare Tunnel — that's the entire serving path for argusweb.bond. The argus_cloudflared container exposes nginx → argus-app/argus-api/phantom/titiler to the public internet.
A git push does NOT redeploy the backend — you must rebuild the container: docker compose up -d --no-deps --build --force-recreate argus-app
Vercel auto-builds from GitHub master but is NOT the backend. It produces a build at a Vercel-hosted preview URL that no live traffic uses. Treat Vercel as a side-effect of pushing, not a deploy target. Don't rely on it; don't tell the user to "check Vercel".
argusweb.space is suspended; ignore it. Active domain is argusweb.bond.
data/settings.json is volume-mounted and intentionally not committed (contains live LLM apiKey). Edit directly — no rebuild needed for settings changes.
Default flow after any code change: build + commit + push + droplet rebuild + verify, all without asking. Use /deploy. Skip only if user says "don't commit" or is mid-iteration.

Do not claim a bug is fixed until you have evidence from the live container — curl the endpoint, tail the logs, or load the page. Show the output.
For LLM/API/feed bugs, hit the route via curl http://localhost/api/... after the rebuild and quote the response.
For UI/Cesium bugs you can't load directly, say so explicitly rather than declaring success.

When the user gives a numbered list, repair plan, or directive ("fix X, Y, Z"), implement directly. Do not invoke brainstorming or ask clarifying questions unless a step is genuinely impossible to start.
When unsure between two reasonable choices on a small detail, make the call and proceed; surface it in the summary.

TypeScript, Next.js 14 (App Router), React, Cesium.js, D3.js
LLM: Gradient (DigitalOcean agent endpoint, OpenAI-compatible) — primary; Ollama optional
Compose stacks: docker-compose.yml (main), docker-compose.realtime.yml (redis), docker-compose.observability.yml (glitchtip). Realtime stack must be up — argus_api depends on Redis.
Run npx tsc --noEmit or npx next build (in argus-app/) after non-trivial changes.

Linux droplet, has Docker, has sudo if needed (but rarely required).
HTTPS via Cloudflare Tunnel, not certbot.
Vercel CLI may not be logged in — prefer the GitHub-push auto-deploy path or the Vercel dashboard.

When removing dead code, grep for ALL references (imports, components, API routes, types, .env files) before committing. Always do a final verification grep.
Run npx next build after removals.