AIWatch
English | 한국어
Real-time monitoring dashboard for 49 AI services — track status, latency, uptime, and incidents across major AI providers.
Dashboard · Landing Page
Share
🛰️ Live Demo
Visit ai-watch.dev — no signup required. Updated every 5 minutes via Cloudflare Workers.
Features
- Real-time status — Operational / Degraded / Down for 49 AI services
- PWA support — Add to home screen, offline cache with Service Worker
- Latency monitoring — Direct endpoint response time (RTT) for 36 probe-capable services; no latency is shown for a service without one
- 24h latency trend — Chart.js line chart with 5-min probe snapshots
- Incident history — Timeline with details from multiple status page formats
- Uptime — 30-day uptime computed by AIWatch from each provider's own published records, rather than copied from the % they display (weighted: full outage 1.0, partial/degraded 0.3, announced maintenance excluded). Sources that don't fit that formula carry their own label instead — so our figure can differ from a provider's own, by design (how it works)
- Component status breakdown — Real-time per-component status (models, API surfaces, …) on ServiceDetails + Is X Down for the services we track per-component, with collapsible section/model groups for long lists
- Status calendar — daily status visualization
- Discord & Slack alerts — Discord webhook on status changes/incidents + Slack via its native
/feed RSS app (zero-config) + RSS feeds
- Cookie consent — GA4 Consent Mode v2 with accept/essential-only
- Deep links — Hash-based routing (
#claude, #latency) for direct page access
- Dark/Light theme — System-aware with manual toggle
- Bilingual — Korean / English
- Mobile responsive — Sidebar overlay, mobile action bar
- AIWatch Score — Composite reliability score combining uptime, incidents, recovery time, and probe-based responsiveness (how it works)
- RTT degradation detection — AIWatch's direct API probes flag latency degradation that official status pages often never report (dashboard badge + Discord daily summary). Independent detection within the ~5-min polling cycle of the official report (MTTD)
- Regional availability — Per-region incident status for SpaceXSI, Gemini, OpenAI with switch recommendation
- Smart alerts — Discord alerts for degraded/down status with anti-flapping, incident suppression, and recovery duration
Monitored Services
Grouped by the dashboard's category taxonomy (49 total — sidebar filters / Overview sections mirror these).
LLM APIs (16)
| Service | Provider | Status Source |
|---|
| Claude API | Anthropic | Atlassian Statuspage |
| OpenAI API | OpenAI | incident.io (Atlassian compat) |
| Gemini API | Google | Google Cloud incidents.json |
| Mistral API | Mistral AI | Rootly (browser-rendered) |
| Cohere API | Cohere | incident.io (Atlassian compat) |
| GroqCloud | Groq | incident.io (Atlassian compat) |
| Together AI | Together | Better Stack RSS + uptime API |
| Fireworks AI | Fireworks | incident.io (Atlassian compat) |
| Cerebras Inference | Cerebras | Atlassian Statuspage |
| Perplexity | Perplexity AI | incident.io (Atlassian compat) |
| SpaceXSI API | SpaceXSI | RSS feed |
| DeepSeek API | DeepSeek | Flashduty (browser-rendered feed) |
| Kimi API | Moonshot AI | Atlassian Statuspage (Chinese titles → English) |
| OpenRouter | OpenRouter | Datadog Status Page (config.json) |
| Amazon Bedrock | AWS | AWS Health Dashboard |
| Azure OpenAI | Microsoft | Azure Status RSS |
Coding Agents (6)
| Service | Provider |
|---|
| Claude Code | Anthropic |
| Codex | OpenAI |
| Cursor | Anysphere |
| GitHub Copilot | GitHub |
| Windsurf (Devin Desktop) | Cognition |
| Junie | JetBrains |
Voice (4)
| Service | Provider | Status Source |
|---|
| ElevenLabs | ElevenLabs | incident.io (Atlassian compat) |
| AssemblyAI | AssemblyAI | Atlassian Statuspage |
| Deepgram | Deepgram | Atlassian Statuspage |
| Fish Audio | Fish Audio | Better Stack RSS + uptime API |
Inference & Infra (9)
| Service | Provider | Status Source |
|---|
| Hugging Face | HuggingFace | Better Stack RSS + uptime API |
| Replicate | Replicate | incident.io (Atlassian compat) |
| fal.ai | fal | Instatus (Next.js) |
| Pinecone | Pinecone | Atlassian Statuspage |
| turbopuffer | turbopuffer | incident.io (Atlassian compat) |
| Voyage AI | Voyage AI | Atlassian Statuspage |
| Jina AI | Jina AI | Atlassian Statuspage |
| Modal | Modal | Better Stack RSS + uptime API |
| Twelve Labs | Twelve Labs | Atlassian Statuspage |
Observability (3)
| Service | Provider | Status Source |
|---|
| LangChain (LangSmith) | LangChain | incident.io (global page) |
| Helicone | Helicone | Better Stack RSS + uptime API |
| Langfuse | Langfuse | incident.io (Atlassian compat) |
Video (2)
| Service | Provider | Status Source |
|---|
| Runway | Runway | Atlassian Statuspage |
| Luma (Dream Machine) | Luma | Better Stack RSS + uptime API |
Image (2)
| Service | Provider | Status Source |
|---|
| Stability AI | Stability AI | incident.io (Atlassian compat) |
| Black Forest Labs (FLUX) | Black Forest Labs | Atlassian Statuspage |
AI Apps (7)
| Service | Provider |
|---|
| claude.ai | Anthropic |
| ChatGPT | OpenAI |
| Character.AI | Character AI |
| DeepSeek App | DeepSeek |
| Grok | SpaceXSI |
| Kimi App | Moonshot AI |
| Grok Bot | SpaceXSI |
Tech Stack
| Layer | Technology |
|---|
| Frontend | React 19, Vite 6, TailwindCSS v4, Chart.js |
| Backend | Cloudflare Workers (TypeScript) |
| Cache | Cloudflare KV (status cache, latency snapshots) |
| Hosting | Vercel |
| Alerts | Discord Webhook (Worker proxy) · Slack via native /feed RSS · RSS feeds |
| Analytics | Google Analytics 4 (Consent Mode v2) |
| Tests | Playwright (E2E), Vitest (unit) |
Architecture
Browser (React SPA, 60s polling)
↓
Cloudflare Worker
├── GET /api/status → parallel fetch (49 services) → normalize
├── GET /api/uptime → daily uptime history
└── POST /api/alert → Discord webhook proxy (SSRF protected)
↓
Parsers (worker/src/parsers/)
├── impact-weights.ts → shared MAJOR_WEIGHT/MINOR_WEIGHT (Atlassian severity formula)
├── uptime-interval.ts → shared trailing-window downtime accumulator for the interval-based parsers (#1006)
├── statuspage.ts → Atlassian Statuspage API + uptimeData HTML (uptime computed per-day)
├── incident-io.ts → incident.io compat API + component_impacts intervals (uptime computed, same weighted formula)
├── gcloud.ts → Google Cloud incidents.json (Vertex Gemini)
├── aistudio.ts → Google AI Studio + direct Gemini API (secondary source, merged with gcloud — #310)
├── instatus.ts → Instatus Nuxt/Next.js SSR
├── betterstack.ts → Better Stack RSS + /index.json uptime API + dailyImpact (status_history)
├── datadog.ts → Datadog Status Page `config.json` (OpenRouter — #1403)
├── flashduty.ts → Flashduty feed (DeepSeek + DeepSeek App — browser-rendered via a scheduled Action, #618)
└── aws.ts → AWS Health events JSON API (Bedrock) + RSS (Azure OpenAI)
↓
Cloudflare KV
├── services:latest (status cache, TTL 5min)
├── daily:YYYY-MM-DD (uptime counters, TTL 2d)
├── history:YYYY-MM-DD (archived counters, TTL 90d)
├── probe:24h (health check probes, max 2016, TTL 7d, 36 probe targets)
├── ai:analysis:{svcId}:{incId} (AI per-incident analysis, TTL 1h, refreshed while active)
├── ai:reanalysis-skip:* (re-analysis failure cooldown, TTL scaled by failure type — #955)
├── ai:usage:{date} (daily AI usage counter, TTL 30d)
├── alerted:* (alert dedup keys, TTL 2h-7d)
├── detected:{svcId} (earliest detection timestamp, TTL 7d)
├── probe-degradation:daily:{svcId}:{date} (RTT degradation counter, TTL 48h, #464)
├── reddit:seen:{postId} (Reddit post dedup, TTL 24h)
└── vitals:{YYYY-MM-DD} (Web Vitals daily aggregation, TTL 3d)
Getting Started
Prerequisites
- Node.js 24 (
.nvmrc)
- npm
- Cloudflare account (for Worker deployment)
Frontend
git clone https://github.com/bentleypark/aiwatch.git
cd aiwatch
npm install
npm run dev # localhost:5173
Worker (Backend)
cd worker && npm install && cd ..
# Create .dev.vars for local dev:
echo "ALLOWED_ORIGIN=*" > worker/.dev.vars
# Run from the repo root — starts the worker on localhost:8788,
# which matches the dashboard's default / VITE_API_URL below.
npm run dev:worker # localhost:8788
Environment Variables
Frontend (.env)
VITE_API_URL=http://localhost:8788/api/status
VITE_GA4_ID= # Optional: Google Analytics measurement ID
Worker (wrangler.toml + secrets)
ALLOWED_ORIGIN=https://your-domain.com
DISCORD_WEBHOOK_URL= # Worker Secret: Discord webhook for alerts
ANTHROPIC_API_KEY= # Worker Secret: Claude Sonnet API key (AI Analysis fallback)
Scripts
# Frontend
npm run dev # Dev server (localhost:5173)
npm run dev:worker # Worker dev server (localhost:8788)
npm run dev:all # Both simultaneously
npm run build # Production build → dist/
npm run lint # ESLint
npm test # Playwright E2E tests
npm run test:src # Frontend unit tests (vitest — incl. CSP hash pin)
npm run test:worker # Worker unit tests (vitest)
# Worker deployment
npm run deploy:worker # Deploy to Cloudflare (use npm script only)
API Endpoints
| Endpoint | Method | Description |
|---|
/api/status | GET | All service statuses + incidents + uptime + aiAnalysis |
/api/status/cached | GET | KV-only cached status (for Edge SSR, fast ~1.2s) |
/api/uptime?days=30 | GET | Daily uptime history (1-90 days) |
/api/report?month=YYYY-MM | GET | Monthly reliability archive (uptime, score, incidents, latency) |
/api/alert | POST | Discord webhook proxy (SSRF protected) |
/badge/:serviceId | GET | SVG status badge (shields.io style) |
/feed.xml | GET | Incident RSS 2.0 — all services (Slack /feed compatible) |
/feed/:slug | GET | Incident RSS 2.0 — single service (slug or service ID) |
/api/og | GET | Dynamic OG image PNG (1200×630, resvg-wasm) |
/api/v1/status | GET | Public API — all services (lightweight, CORS *) |
/api/v1/status/:id | GET | Public API — single service + top 5 incidents |
Public API (v1)
Open API for external developers. No authentication required. Rate limited to 60 req/min.
All services:
curl https://aiwatch-worker.p2c2kbf.workers.dev/api/v1/status
Single service:
curl https://aiwatch-worker.p2c2kbf.workers.dev/api/v1/status/claude
Response includes: id, name, provider, category, group (fine taxonomy — llm / voice / inference / …), status, latency (the service's latest direct-probe RTT in ms; null when it is not probed — never status-page timing, #1633), uptime30d, uptimeSource, lastChecked, and up to 5 recent incidents (single service only).
Status Badges
Embed real-time status badges in your README, docs, or blog.
[](https://ai-watch.dev/is-claude-down)
Parameters
| Parameter | Description | Example |
|---|
uptime | Show uptime % | /badge/claude?uptime=true |
style | flat or flat-square | /badge/claude?style=flat-square |
label | Custom label | /badge/claude?label=My+API |
Examples
Available Service IDs
Every monitored service — the same ids /api/v1/status returns.
| ID | Service | ID | Service |
|---|
claude | Claude API | elevenlabs | ElevenLabs |
openai | OpenAI API | assemblyai | AssemblyAI |
gemini | Gemini API | deepgram | Deepgram |
bedrock | Amazon Bedrock | huggingface | Hugging Face |
azureopenai | Azure OpenAI | replicate | Replicate |
mistral | Mistral API | fal | fal.ai |
cohere | Cohere API | modal | Modal |
groq | GroqCloud | voyageai | Voyage AI |
together | Together AI | pinecone | Pinecone |
fireworks | Fireworks AI | turbopuffer | turbopuffer |
cerebras | Cerebras Inference | twelvelabs | Twelve Labs |
perplexity | Perplexity | langsmith | LangChain (LangSmith) |
xai | SpaceXSI API | helicone | Helicone |
deepseek | DeepSeek API | langfuse | Langfuse |
kimi | Kimi API | runway | Runway |
openrouter | OpenRouter | luma | Luma (Dream Machine) |
claudecode | Claude Code | stability | Stability AI |
|
Claude Code Statusline Integration
Surface AI service outages — Claude API, OpenAI, Gemini, GitHub Copilot, and 45 more — directly in your Claude Code statusline. The recommended preset keeps an always-on, clickable AIWatch label (AIWatch 🟢 while all healthy, AIWatch 🔴 Claude API when something breaks — cmd/ctrl+click opens the dashboard). Prefer zero footprint when healthy? A minimalist preset that stays empty until something degrades is on the presets page.
Quickest install — add to ~/.claude/settings.json:
{
"statusLine": {
"type": "command",
"command": "( curl -sf --max-time 2 https://aiwatch-worker.p2c2kbf.workers.dev/api/statusline/branded ) 2>/dev/null || true"
}
}
Requires an OSC 8-compatible terminal (iTerm2, Warp, kitty, WezTerm, VS Code terminal, Terminal.app on macOS 12+) for the clickable label; others show it as plain text. Targets the Worker domain directly so per-prompt polls don't count as Vercel bandwidth.
Other presets — minimalist (empty when healthy), compact badge, full list, scoped to specific providers, and clickable per-service links: ai-watch.dev/#statusline
Properties: single GET per render, 5-min KV-cached on Cloudflare's edge, 2-second timeout, no jq dependency, no Anthropic API requests, no client identifier. On a network error the line stays empty; when AIWatch cannot read its own status snapshot it renders AIWatch ⚪ (unknown) rather than a green it cannot support. Compatible with any statusline tool that supports shell-command output (including ccstatusline's Custom Command widget).
Claude Code Plugin (Beta)
Beyond the status bar, the AIWatch Claude Code plugin surfaces AI outages inside Claude Code with two pieces:
- Background outage monitor — notifies you the moment a monitored provider goes down (
🔴 Claude API is down) or recovers (✅ Claude API has recovered), naming each service. Diffs each poll (default every 60s), so it only speaks up on a real transition — never spams.
/aiwatch command — briefs which AI services are degraded/down right now, each with its active incident (title + impact), an AI summary, and a fallback suggestion.
Reads no code and collects no data — it only polls AIWatch's public status feed. Install it from AIWatch's own marketplace:
/plugin marketplace add bentleypark/aiwatch
/plugin install aiwatch@aiwatch-dev
Source + docs: plugin/aiwatch/. Full details: ai-watch.dev/plugin.
Project Structure
src/ # React 19 SPA (Vite, no router — hash routing in App.jsx)
components/ # Shared UI: StatusPill, SkeletonUI, EmptyState, Modal, Sidebar, Topbar, CookieBanner, AnalysisModal, …
pages/ # Overview, Latency, Incidents, Uptime, ServiceDetails, Settings, Ranking, Statusline
hooks/ # usePolling, useTheme, useLang, useSettings, useGitHubStars, useMonthlyArchives
utils/ # analytics, calendar, time, pageContext, constants, hashRoute, …
locales/ # ko.js, en.js (flat key→string maps)
api/ # Helpers live in `_`-prefixed dirs and handlers run on the edge runtime,
# so neither counts against the Hobby 12-Serverless-Function cap (#862/#867)
intro.ts # Landing page (/intro) _intro/ # its SSR template
is-down.ts # "Is X Down?" SSR pages (47 services) _is-down/ # slug-map, seo-content, template
methodology.ts # "How AIWatch Works" (/methodology) _methodology/
plugin.ts # Claude Code plugin landing _plugin/
badges.ts # Status-badge gallery (/badges) _badges/
reports.ts # /reports/* proxy → Jekyll site _shared/ # shared Edge helpers
confirm.ts csp-report.ts plugin-privacy.ts extension-privacy.ts
public/
manifest.json # PWA manifest (the Service Worker is built from src/sw/ → /sw.js)
icon-192.png # PWA icon 192x192
icon-512.png # PWA icon 512x512
scripts/ # Build/CI/ops scripts (OG generation, CI lint gates, verify-reminders, …)
worker/
src/
index.ts # Worker entry: CORS, routing, /api/*, /badge, /feed, cron scheduled handler
services.ts # Service configs + fetch orchestrator + status determination
types.ts utils.ts # Shared types + shared utilities
score.ts # AIWatch Score calculation
alerts.ts # Alert detection (incident + status-edge), holds, merges
ai-analysis.ts # Hybrid AI incident analysis (Gemma 4 primary + Sonnet fallback)
anthropic.ts # Anthropic Messages REST call (model id, retry, status classification)
probe.ts # Health check probing — direct RTT measurement (`PROBE_TARGETS`)
rss.ts badge.ts og.ts og-render.ts # Feeds, badges, OG images
daily-summary.ts weekly-briefing.ts monthly-archive.ts monthly-narrative.ts
security-monitor.ts # AI service security monitoring (HN Algolia, OSV.dev SDK vulnerabilities — 24 tracked packages)
changelog.ts reddit.ts platform-monitor.ts
suppression.ts overrides.ts withdrawn.ts upstream-link.ts # Operator + incident layers
parsers/ # Platform-specific parsers
statuspage.ts # Atlassian Statuspage
incident-io.ts # incident.io (Atlassian-compatible API)
gcloud.ts # Google Cloud Vertex (gemini primary source)
aistudio.ts # Google AI Studio + direct Gemini API (gemini secondary, #310)
instatus.ts # Instatus
betterstack.ts # Better Stack
datadog.ts # Datadog Status Page (OpenRouter)
flashduty.ts # Flashduty (DeepSeek + DeepSeek App)
aws.ts # AWS Health events JSON API — Bedrock (+ RSS parser reused for Azure OpenAI)
impact-weights.ts uptime-interval.ts # Shared uptime primitives (#1006)
__tests__/ parsers/__tests__/ # Vitest unit tests
This is a map, not an inventory — worker/src/ has ~50 modules. The per-module purpose and the
issues that shaped each one live in docs/reference/directory-map.md.
Contributing
See CONTRIBUTING.md for detailed guidelines.
All coding agents should start with AGENTS.md and the
agent-neutral development workflow. Claude Code
contributors should additionally read CLAUDE.md for Claude-specific automation rules.
- Fork the repository
- Create a feature branch (
git checkout -b feature/my-feature)
- Follow the shared workflow in AGENTS.md and
docs/reference/development-workflow.md; Claude Code
contributors should also follow CLAUDE.md
- Build + test:
npm run build && npm test && npm run test:src && npm run test:worker
- Submit a pull request using the PR template
Issues
Pull Requests
- One feature or fix per PR
- All tests must pass (
npm test + npm run test:src + npm run test:worker) — CI-gated, with E2E running on frontend changes
- Include
closes #N in commit messages
- Fill out the PR checklist
Security
Found a vulnerability? Please report it responsibly — see SECURITY.md for details.
License
AGPL-3.0 covers the code. The AIWatch name and logo are not licensed with it; see the Trademark Policy. The official site is ai-watch.dev.
