diff --git a/site/app.js b/site/app.js
new file mode 100644
index 0000000..2fa0d85
--- /dev/null
+++ b/site/app.js
@@ -0,0 +1,245 @@
+// status.moleculesai.app — read-only status page for Molecules AI services.
+//
+// Pulls the probe-list config + per-site history JSONL from the
+// molecule-ai-status repo on Gitea, renders a one-row-per-service
+// dashboard with current state + a 24h-history sparkline.
+//
+// Why no framework: this page is plain DOM + fetch. Zero build step,
+// zero dependencies, zero supply-chain surface. The thing it MUST do
+// well is "load fast, show correct status, never lie." React/Vue
+// would be cargo-culting at this scale.
+//
+// Data source: public Gitea raw URLs. No auth. Repo is public.
+
+const REPO_BASE =
+ "https://git.moleculesai.app/molecule-ai/molecule-ai-status/raw/branch/main";
+const HISTORY_URL = (slug) => `${REPO_BASE}/history/${slug}.jsonl`;
+const CONFIG_URL = `${REPO_BASE}/.upptimerc.yml`;
+
+// Window of history we render in the sparkline (24h of probes at one
+// per 5 minutes ≈ 288). Cap to keep the DOM bounded if a site has
+// been probing for years.
+const SPARKLINE_LIMIT = 288;
+
+// Slugify must match the probe binary's slugify() in cmd/probe/main.go
+// — the page reads files the probe writes, so the slugging rule is
+// load-bearing. Mirror in tests if/when this gets a follow-up.
+function slugify(s) {
+ let out = "";
+ let last = "-";
+ for (const c of s.toLowerCase()) {
+ const isAlnum = (c >= "a" && c <= "z") || (c >= "0" && c <= "9");
+ if (isAlnum) {
+ out += c;
+ last = c;
+ } else if (last !== "-") {
+ out += "-";
+ last = "-";
+ }
+ }
+ return out.replace(/^-+|-+$/g, "");
+}
+
+// Minimal YAML parser for the subset of .upptimerc.yml we read:
+// only the `sites:` list of `{name, url}`. Anything more elaborate
+// (anchors, multiline strings, etc.) is overkill — the upstream
+// upptime config schema is intentionally simple.
+function parseSites(yamlText) {
+ const sites = [];
+ let inSites = false;
+ let current = null;
+ for (const rawLine of yamlText.split("\n")) {
+ const line = rawLine.replace(/\r$/, "");
+ if (line.startsWith("#")) continue;
+ if (/^\s*$/.test(line)) continue;
+
+ if (/^sites:\s*$/.test(line)) {
+ inSites = true;
+ continue;
+ }
+ if (inSites && /^[a-zA-Z]/.test(line)) {
+ // hit a top-level key after sites: — bail
+ inSites = false;
+ }
+ if (!inSites) continue;
+
+ const itemStart = line.match(/^\s*-\s+name:\s*(.+)$/);
+ if (itemStart) {
+ if (current) sites.push(current);
+ current = { name: itemStart[1].trim().replace(/^["']|["']$/g, "") };
+ continue;
+ }
+ const urlMatch = line.match(/^\s+url:\s*(.+)$/);
+ if (urlMatch && current) {
+ current.url = urlMatch[1].trim().replace(/^["']|["']$/g, "");
+ }
+ }
+ if (current) sites.push(current);
+ return sites.filter((s) => s.name && s.url);
+}
+
+// Parse a JSONL response into an array of Result objects. Tolerant of
+// trailing newlines + (rarely) blank lines from a partial-write race.
+function parseJSONL(text) {
+ const out = [];
+ for (const line of text.split("\n")) {
+ if (!line.trim()) continue;
+ try {
+ out.push(JSON.parse(line));
+ } catch {
+ // skip malformed line — better than the whole page erroring
+ }
+ }
+ return out;
+}
+
+// Best-effort fetch — returns null on failure (no exceptions).
+async function fetchText(url) {
+ try {
+ const resp = await fetch(url, { cache: "no-cache" });
+ if (!resp.ok) return null;
+ return await resp.text();
+ } catch {
+ return null;
+ }
+}
+
+// Render a row for one site given its latest results.
+function renderRow(site, results) {
+ const last = results[results.length - 1];
+ const status = !last
+ ? "unknown"
+ : last.success
+ ? "up"
+ : "down";
+
+ const latency = last && last.success ? `${last.latency_ms} ms` : "—";
+
+ // Sparkline: last SPARKLINE_LIMIT entries, one bar per. Bar height
+ // proportional to latency (clamped). Failing checks render red and
+ // taller (so eye is drawn to outages).
+ const recent = results.slice(-SPARKLINE_LIMIT);
+ const maxLat = Math.max(50, ...recent.filter((r) => r.success).map((r) => r.latency_ms));
+
+ const spark = recent
+ .map((r) => {
+ const cls = r.success ? "" : "fail";
+ const h = !r.success ? 20 : Math.max(2, Math.round((r.latency_ms / maxLat) * 18));
+ return ``;
+ })
+ .join("");
+
+ return `
+
+
+
+
${spark}
+
${latency}
+
+ ${text}
+ ${sub}
+
+ `;
+}
+
+async function load() {
+ // 1. Fetch + parse the probe-list config.
+ const yaml = await fetchText(CONFIG_URL);
+ if (!yaml) {
+ document.getElementById("grid").innerHTML =
+ `Failed to load probe-list config from Gitea. Check that ${CONFIG_URL} is reachable.
`;
+ document.getElementById("updated").textContent = "load failed";
+ return;
+ }
+ const sites = parseSites(yaml);
+ if (sites.length === 0) {
+ document.getElementById("grid").innerHTML =
+ `No sites declared in .upptimerc.yml.
`;
+ return;
+ }
+
+ // 2. For each site, fetch its history JSONL in parallel.
+ const enriched = await Promise.all(
+ sites.map(async (site) => {
+ const slug = slugify(site.name);
+ const text = await fetchText(HISTORY_URL(slug));
+ const results = text ? parseJSONL(text) : [];
+ return { site, slug, results };
+ })
+ );
+
+ // 3. Render rows + summary.
+ const rowSummaries = enriched.map(({ site, results }) => {
+ const last = results[results.length - 1];
+ return {
+ status: !last ? "unknown" : last.success ? "up" : "down",
+ };
+ });
+
+ document.getElementById("summary").innerHTML = renderSummary(rowSummaries);
+ document.getElementById("grid").innerHTML = enriched
+ .map(({ site, results }) => renderRow(site, results))
+ .join("");
+
+ // Updated-at timestamp: latest probe across all sites.
+ const allTimestamps = enriched
+ .flatMap(({ results }) => results)
+ .map((r) => r.timestamp)
+ .filter(Boolean);
+ if (allTimestamps.length > 0) {
+ const latest = allTimestamps.sort().pop();
+ const ago = Math.round((Date.now() - new Date(latest).getTime()) / 60000);
+ document.getElementById("updated").innerHTML =
+ `last probe ${ago} min ago · history`;
+ } else {
+ document.getElementById("updated").textContent = "no probe data yet";
+ }
+}
+
+load();
+// Auto-refresh every 5 min — matches the probe cadence so the page
+// catches up with new history without a hard reload.
+setInterval(load, 5 * 60 * 1000);
diff --git a/site/index.html b/site/index.html
new file mode 100644
index 0000000..d06e278
--- /dev/null
+++ b/site/index.html
@@ -0,0 +1,137 @@
+
+
+
+
+
+
+Molecules AI · Status
+
+
+
+
+
+ Molecules AI · Status
+ checking…
+
+
+
+
+
+ Loading current status…
+ Fetching latest probe results.
+
+
+
+
+
+
+