Changelog

What's new.

Subscribe to the RSS feed or follow @teravor.

May 5, 2026

v0.10.0

Feature

v0.10.0

Dashboard status-page list view: stat cards (all/published/draft counts with plan-limit display), search by title or slug, published/draft filter bar, grid table with brand-color dot, slug, component count, status chip, relative updated time, View and Delete actions. Create button is plan-limit-gated (disabled with upgrade CTA when at limit). Removes the single-page UI guard that skipped the list when no pages existed (T-230).
Status page uptime percentage no longer double-scales the value (was multiplying by 100 a second time after the `pct()` helper already returned 0–100).

May 5, 2026

v0.9.4

Feature

v0.9.4

Focus trap on hand-built dialog components (Add Monitor, channel form, shortcuts, billing confirm) for keyboard accessibility.
`loading.tsx` skeleton fallbacks for app, marketing, admin, billing, and monitors route segments — no more blank screens during navigation.
ARIA tab roles on the monitor detail tab bar (`role="tablist"`, `role="tab"`, `aria-selected`, `aria-controls`, `role="tabpanel"`).
`aria-label` and `role="status"` on `StatusDot` so screen readers announce monitor state.
New migration: partial index on `check_results(monitor_id, checked_at DESC) WHERE status = 'down'` for heartbeat-sweep performance.
New migration: 90-day retention on `whop_webhook_events` via `pg_cron` + `prune_whop_webhook_events()`.
New migration: 30-day retention on dead-lettered `notification_attempts` via `pg_cron`.
New migration: index on `notification_attempts(incident_id, event, channel_id)` for delivery-summary GROUP BY.
Go probe: config validation tests, SSL and ICMP monitor config tests.
`BETTER_AUTH_SECRET` now rejects the known development default when `NODE_ENV=production`, preventing trivially forgeable session tokens.
`WHOP_PLANS_JSON` now warns when empty in production — an empty config previously caused every new subscription to silently fall back to the `'pro'` tier.
Go probe: DNS checker resolves the resolver hostname once and dials the IP directly, closing a DNS-rebinding TOCTOU that could bypass netguard on the second resolution.
Go probe: `processBatch` now caps concurrent goroutines with a semaphore (max 50), preventing resource exhaustion at large `BatchSize` values.
Go probe: dial timeout derived from per-check timeout (capped at 5 s) instead of hardcoded 10 s.
Go probe: `BatchSize` clamped to max 100, `PollInterval` floored at 50 ms, `PROBE_REGION` validated as alphanumeric on startup.
Go probe: HTTP URL validation rejects empty hosts and URLs with embedded credentials; TCP `send` capped at 10 KB; DNS `resolver` validated as `host:port`.
Go probe: ICMP echo ID uses `crypto/rand` instead of PID, reply buffer allocated once before the inner loop, body-match uses `bytes`/`[]byte` instead of `string` to halve peak memory.
Status page uptime stats now query the `check_results_daily` CAGG instead of the raw hypertable (~90 rows vs ~129 k per monitor).
Admin org detail endpoint runs 4 independent queries in `Promise.all` instead of sequentially.
Heartbeat sweep uses a single batch query for stale-monitor "recent down" checks instead of N+1 per-monitor queries.
Duplicated advisory-lock acquisition consolidated into a single `acquireAdvisoryLock()` in `lib/leader.ts`; notifiers import the shared version.
Duplicated `toIso` helper extracted to shared `lib/dates.ts` from three service files.
Shared `API_INTERNAL_URL` constant in `lib/config.ts` replaces three separate definitions.
`alert-channels` PATCH config check uses `!== undefined` instead of truthiness, fixing validation skip on `{}`.
IP rate-limiter and waitlist rate-limiter Maps now periodically evict stale entries (60 s interval, unref'd).
Invoice cache now periodically evicts expired entries.
Signup page Terms and Privacy links pointed at `/legal/terms` and `/legal/privacy` (404s); now correctly link to `/terms` and `/privacy`.
`sql.lit()` for user-supplied tag parameter replaced with parameterized `ARRAY[${tag}]::text[]`.
`sql.raw()` in dead-letter replay now guarded by an explicit allowlist check.
Invite redemption token restoration now runs in a `db.transaction()` on signup failure, preventing a TOCTOU broken-invite state.
Waitlist, admin-waitlist, and invite error responses now throw `HTTPException` consistently instead of returning ad-hoc JSON shapes (some missing `message` field).
Incident `scope` query parameter now validated through the Zod schema instead of raw cast.
`io.ReadAll` error in Go HTTP checker now produces a proper down result instead of a false body-match failure.
`downResult` in Go HTTP checker now guards against nil error to prevent a panic.
Go probe handlers map now protected by `sync.RWMutex` with a `started` guard to prevent data races if `Register` is called after `Run`.
Go probe `SetDeadline` errors now logged instead of silently discarded.
Empty catch blocks in ownership verification now log at debug level for diagnosability.
`console.error` in pool error handlers replaced with pino structured logger.
Grace-reminder processor now logs a warning when `downgrade_grace_until` is set but `pending_plan` is null (data inconsistency).
Swallowed `api_key.last_used_at` update error now logged with `logger.warn`.
`apiFetch` now throws `ApiError` when the response body contains an `{error}` key, catching mismatched error shapes at the boundary.
`add-monitor-flow` draft state typed as `MonitorDraft` instead of `any`.
`team-billing` `PLANS.find()` non-null assertions replaced with safe fallbacks.
`status-page-editor` `window.confirm()` replaced with the project's `AlertDialog` component.
Non-functional subscribe forms on status pages replaced with "coming soon" text.
Toast timeout now cleared on unmount to prevent `setState` on unmounted component.
Docs scroll spy throttled with `requestAnimationFrame` to avoid expensive per-pixel DOM reads.
Footer copyright year is now dynamic instead of hardcoded `2026`.
`.env.example` Whop sandbox keys replaced with placeholder values.

May 5, 2026

v0.9.3

Fix

v0.9.3

Caddy no longer marks the entire API upstream as unhealthy when a single route returns 5xx — passive health checks now only trigger on connection failures.

May 5, 2026

v0.9.2

Fix

v0.9.2

Zero-downtime deployments: Caddy passive health checks, API drain on shutdown, and warmup timing eliminate 521 errors during container rollover.
`PATCH /api/monitors/:id` config validation now validates against the specific monitor type's schema instead of the full union, preventing false rejections of valid edits.

May 4, 2026

v0.9.1

Fix

v0.9.1

API route validation schemas now use typed Zod unions (`httpConfigSchema`, `emailConfig`, `brandingSchema`, etc.) instead of `z.unknown()`, improving OpenAPI spec accuracy and request/response validation for monitors, alert channels, and status pages.

May 4, 2026

v0.9.0

Feature

v0.9.0

`/admin/ops` now shows a **Cron jobs · pg_cron** card listing the last 5 runs per scheduled job, including status, return message, start time, and duration. The new `GET /api/admin/ops/cron-runs` endpoint queries `cron.job_run_details` and emits a Sentry warning when the most recent run for any `argus-*` job has `status = 'failed'`.
`/admin/ops` queues card now shows a yellow `LAGGY` chip when `oldest_msg_age_sec > 60` and a red `STUCK` chip when `> 300`, so queue health is visible at a glance.
API emits a Sentry `warning` when a PGMQ queue crosses the LAGGY or STUCK threshold, deduped per-queue per-15-minutes to avoid alert fatigue.
**Sentry release tagging:** `SENTRY_RELEASE` env var is now wired into the API and probe Sentry init blocks. `scripts/deploy.sh` passes `TAG` as `SENTRY_RELEASE` so every deployed image is tagged with its commit SHA in Sentry, enabling accurate "issue introduced in" timelines and per-release regression tracking.
Security page at `/security` with accurate infrastructure details: Hetzner Falkenstein hosting, Caddy TLS, bcrypt/Better Auth, SSH key-only access, and responsible disclosure contact. Footer link re-enabled.
Bulk monitor operations: select multiple monitors on the dashboard with per-row checkboxes and a "select all" header checkbox; a floating action bar offers Pause, Resume, and Delete (with confirmation dialog) for all selected monitors at once.
API endpoints `POST /api/monitors/bulk/pause`, `/bulk/resume`, and `/bulk/delete` — cap at 100 IDs per request, scope-checked against the requesting org, return `{ affected: number }`.
Editing a monitor in the Settings tab no longer gets wiped by background 30-second refetches. The form now tracks a `dirty` flag and skips the prop-to-state sync while the user has unsaved changes; syncing resumes after a successful save.
Replaced array-index React keys with stable values in the monitor assertions Chip lists (`monitor-assertions.tsx`) and the plan feature list in the billing page (`team-billing.tsx`), preventing spurious remounts when lists update.
Dashboard topbar: removed non-functional search input, "Public status ↗" placeholder link, and notifications bell (all were decorative with no implementation).

May 1, 2026

v0.8.2

Fix

v0.8.2

The 60-day/90-day uptime bar on the dashboard rendered the rightmost ~1 day as `unknown` (gray) even when probe data was up-to-date. The `check_results_daily` continuous aggregate added in v0.8.0 inherited TimescaleDB's post-2.13 default of `materialized_only = true`, so reads saw only the materialized portion — and the refresh policy can't materialize today's bucket (the bucket only completes after day boundary + the 1h `end_offset`). New migration enables realtime mode (`materialized_only = false`); reads now union the materialized rollup with a real-time computation against the raw hypertable for any bucket past the watermark.

May 1, 2026

v0.8.1

Fix

v0.8.1

Deploy migration `1776988800046_check_results_indexes.sql` failed against production because Timescale hypertables don't support `CREATE INDEX CONCURRENTLY` (the planner propagates the index to all chunks in one transaction; node-pg-migrate also runs each migration inside a transaction). Switched to a plain `CREATE INDEX IF NOT EXISTS` so the next deploy applies cleanly. The v0.8.0 deploy job that errored on this migration left no `pgmigrations` row, so re-running picks up where it stopped.

Apr 30, 2026

v0.8.0

Feature

v0.8.0

Sentry distributed tracing on API HTTP requests and probe operations, sampled via the new `SENTRY_TRACES_SAMPLE_RATE` env var (`0` disables; `0.1` is the prod default in the docs). API transactions are named after the matched Hono route pattern so `/api/monitors/:id` doesn't explode into one transaction per id. Slow endpoints now show up as span chains in Sentry Performance instead of being invisible.
Tick errors in every batch processor (scheduler, alert engine, four notifiers, overage processor, grace-reminder processor) now reach Sentry with a `module` tag, instead of being logged and dropped. The shared runner captures once at the catch-all so the per-processor code stays clean; the scheduler's heartbeat-write swallow gets the same treatment so a sustained outage of the ops-metrics row shows up in Sentry alongside the STALE chip on `/admin/ops`.
Explicit Postgres connection-pool sizing on the API (`DB_POOL_MAX`, `DB_POOL_MIN`, `DB_POOL_IDLE_MS`, `DB_POOL_CONNECTION_TIMEOUT_MS`). Defaults to a conservative 10-connection cap that fits a single API replica plus the scheduler, four notifiers, and Better Auth pool against a small Postgres instance. Pools no longer crash the process on a transient idle-client drop.
TimescaleDB compression and retention policies on `check_results`: chunks older than 7 days are compressed (segmented by `monitor_id`, ordered by `checked_at DESC`) and chunks older than 100 days are dropped. The dashboard hits a 90-day window; 10 days of buffer protects boundary skew. Storage growth on the time-series table is now bounded.
Daily retention sweep on `notification_attempts` via `pg_cron`: the new `prune_notification_attempts(retain_days)` function deletes rows older than 60 days every night at 03:30 UTC. The Alerts tab on monitor detail joins this table for delivery status, but the dashboard's incident window only reaches back 30 days, so 60-day retention keeps every queryable incident's delivery trail intact while bounding table growth.
TimescaleDB continuous aggregate `check_results_daily` rolls up per-monitor per-day check counts (`n`, `up_count`, `bad_count`). Refresh policy slides the last 7 days forward every 30 minutes with a 1-hour end-offset; realtime mode keeps the un-materialized last hour fresh against the raw table. Sets up the foundation for the next stats-query rewrite to read aggregates instead of scanning raw time-series.
Admin **Dead letters** panel on `/admin/ops` — surfaces every dead-lettered notification (retry-exhaustion or channel-type mismatch) with queue, monitor, attempt count, last error, and archived time. New `GET /api/admin/ops/dead-letters` endpoint backs it. Operators no longer have to query PGMQ archive tables by hand to find what failed.
**Replay** button on each Dead-letters row (`POST /api/admin/ops/dead-letters/:id/replay`) re-enqueues the original message onto its source queue with the attempt counter reset, leaves a `replayed` audit row in `notification_attempts`, and is idempotent under double-clicks. Returns 410 when the archived payload has aged out and 409 when it has already been replayed.
Improved mobile responsiveness across landing page and app dashboard: collapsible sidebar drawer with backdrop overlay (hamburger button in topbar), responsive landing nav with mobile hamburger menu, all landing section grids adapt from single column on phones through desktop multi-column layouts, section padding scales with `clamp()`, monitors table wraps in a horizontal-scroll container, 90-day uptime bar scrolls horizontally instead of overflowing, stat card grids collapse to 2 columns on small screens, Add Monitor modal caps at 92 vw on narrow viewports (T-216).
Standardized empty- and error-state surfaces across the dashboard. New shared `<EmptyCard>` and `<ErrorCard>` primitives replace ad-hoc inline strings on the monitor detail (load + stats), response log, alerts tab, monitors list, and alert-channels list. Errors are sanitized: `ApiError` 401/403/404/429/5xx and network failures get user-friendly copy, with the raw message tucked inside a collapsible "Technical details" block instead of dumped inline.
Dashboard detail-page polling now pauses while the tab is hidden. The four monitor query hooks (`useMonitorStatsQuery`, `useMonitorChecksQuery`, `useMonitorAlertEventsQuery`, and the list-view `useMonitorsStatsQuery`) now set `refetchIntervalInBackground: false`; refetches resume the moment the tab regains focus.
Monitor dashboard's 90-day uptime bar (and the list view's 60-day variant) now read from the `check_results_daily` continuous aggregate instead of scanning raw `check_results`. The dashboard pulls ~90 pre-aggregated rows per monitor on each poll instead of grouping over ~129k raw rows for a 60s-interval monitor. Sub-day-granularity queries (latest checks, sparkline, 24h uptime/latency) still read the raw table because they only touch one or two recent chunks.
Tagged releases now deploy automatically. Pushing a `v*` tag (typically via `scripts/release.sh`) triggers `release.yml`, which builds the three images, pushes them to GHCR, and then SSHes into the Hetzner host to run the existing `scripts/deploy.sh`. The laptop no longer needs SSH access for routine deploys; `scripts/release.sh` just bumps the version, promotes the CHANGELOG, pushes the tag, and watches CI. Manual `scripts/deploy.sh` remains supported for emergencies.
Alert engine: a `closed` alert against an incident that is still open is now suppressed with reason `incident_still_open` instead of being fired. The previous behavior was asymmetric — `opened` against an already-closed incident was guarded, but `closed` against a still-open incident was not — which meant a stale or replayed `closed` payload could trigger a recovery email on a live outage.
Probe consumer: a check that fails with the same error each retry no longer thrashes the queue at the static visibility-timeout cadence. The consumer now pushes each retry past an exponential backoff (`PROBE_BACKOFF_BASE_SEC`, default 30s, doubling to `PROBE_BACKOFF_MAX_SEC`, default 600s, ±20% jitter) and archives the message proactively on the final attempt instead of burning one more VT cycle just to detect "still failing".
Heartbeat scheduler sweep: dropped the `COALESCE(last_ping_at, created_at)` fallback that compensated for never-pinged heartbeat monitors. Going forward, `createMonitor` seeds `last_ping_at` to creation time and a CHECK constraint enforces non-null on heartbeat type, so the sweep query and its index can run on a single column. Behaviour is unchanged from the operator's perspective — a heartbeat monitor opens an incident at exactly `interval + grace` after the last (or seeded) ping.
Notifiers: an alert routed to a channel of the wrong type (e.g. an email payload pointing at a Slack channel id, possible only on schema drift or alert-engine fanout bugs) is now dead-lettered with a typed `notification_attempts` row and a Sentry exception. Previously such payloads were silently archived with no record, hiding the fanout bug behind apparent silence.
CI: alert-engine processor tests and the Go probe round-trip seeder now set `last_ping_at` when inserting heartbeat monitors directly, so they satisfy the `monitors_heartbeat_last_ping_at_check` CHECK constraint added with the heartbeat-sweep cleanup. (Previously both bypassed the service-layer `createMonitor()` helper that already seeded the column, and the constraint refused the insert at test time.)

Apr 21, 2026

v0.7.0

Feature

v0.7.0

Self-hosted mailbox on `mail.teravor.com`: receive at role addresses (`admin@`, `alerts@`, `abuse@`, `postmaster@`, `hello@`, `support@`, `contact@`) and a catch-all — all delivering to one IMAP inbox. Powered by Stalwart Mail Server (replaces Postfix).
Contact page at `/contact` with support and security-report email addresses.

Apr 21, 2026

v0.6.0

Feature

v0.6.0

Interval floor enforcement in the add-monitor and monitor-settings forms: interval buttons below the plan minimum are disabled with upgrade tooltips, and the add-monitor flow defaults to the fastest interval the current plan allows.
Monitor count gate in the add-monitor flow: a banner on step 3 shows the limit and an upgrade link when the org is at capacity; the Create button is disabled until the limit is lifted.
Sidebar plan label, nav counts (monitors, status pages, integrations), and the monitors-list stat card are now driven by live API data instead of hardcoded values.
Hardcoded `monitor.teravor.com` domain in the status-page editor, status embed widget, and public status page route replaced with `NEXT_PUBLIC_APP_URL`.

Apr 21, 2026

v0.5.1

Fix

v0.5.1

Production email delivery: the Postfix relay advertised STARTTLS without a server cert, so nodemailer auto-negotiated TLS and every send failed with `454 4.7.0 TLS not available due to local problem` (all notifications dead-lettered). Switched the internal :587 relay to `smtpd_tls_security_level=none` since traffic stays on the Docker bridge; the outbound hop to remote MXes still uses opportunistic TLS. Also swapped the flaky `nc`-based healthcheck for `postfix status`.
Postfix SASL credentials were silently empty in production, causing `454 4.7.0 Temporary authentication failure: generic failure` on every notifier send. Two layered bugs: compose-time `${VAR}` substitution reads `.env` from the compose file's directory (`docker/`), not the repo root, so `${POSTFIX_SMTP_USER:-}` expanded to `""`; and boky/postfix provisions its SASL account from `SMTPD_SASL_USERS` (format `user:pass`), not `SMTP_USERNAME`/`SMTP_PASSWORD` (those are for outbound *relay* auth, unused here). The service now uses `env_file: ../.env` plus `POSTFIX_smtpd_sasl_local_domain=teravor.com` so nodemailer's bare-username SASL login matches the `[email protected]` entry in sasldb. `.env.example` and `docs/email-notifier.md` updated.

Apr 20, 2026

v0.5.0

Feature

v0.5.0

Monitor detail page: the **Response Log**, **Assertions**, and **Alerts** tabs are now implemented (previously stubs). Response Log paginates through raw check results newest-first. Assertions renders a read-only view of the monitor's configured assertions (expected status codes, body match, DNS records, SSL expiry thresholds, TCP expect/send, heartbeat grace) alongside a recent-failures list. Alerts shows assigned delivery channels plus an alert-engine decision history with per-channel delivery status.
API: `GET /api/monitors/:id/checks` and `GET /api/monitors/:id/alert-events` — paginated (ISO-timestamp `before` cursor) endpoints backing the new tabs.
Monitor **Settings** tab is now a full edit surface: every type-specific config field (URL, method, expected status, body match, DNS record/values, SSL warn/critical days, TCP host/port/send/expect, ICMP count/packet size, heartbeat grace) plus check interval, enabled state, friendly name, and alert channels can be edited in place. Reuses the same type-specific forms as the Add Monitor flow.
Monitor detail Overview: the disabled "Full response log →" button now navigates to the Response Log tab, and the "Assertions" placeholder card becomes a nav card to the Assertions and Alerts tabs.
API: `PATCH /api/monitors/:id` now accepts a typed `config` body (validated per monitor type via the same Zod schemas as create). The monitor `type` remains immutable — changing type still requires delete + create.
API: when a config edit moves a raw-IP monitor to a new host, the monitor is disabled and the response carries a fresh ownership challenge (`token`, `well_known_url`, `dns_name`); the caller must re-run `POST /api/monitors/:id/verify-ownership` before probes resume.

Apr 20, 2026

v0.4.3

Feature

v0.4.3

Monitor detail "Edit" button is now functional — jumps to the Settings tab, which gained an editable Friendly name field. Save persists name and channel changes together, and works even when no alert channels are configured.
Add Monitor flow auto-derives a friendly name from the monitor config (HTTP hostname, `host` for DNS/TCP/SSL/ICMP, dated label for heartbeats) when the user leaves the field blank, instead of the monitor surfacing as a UUID slice.
Monitors list falls back to the config `host` (not the id slice) when a monitor has no name, so DNS/TCP/SSL/ICMP monitors display sensibly without a name.
Monitors list 60-day uptime bar now actually shows the colored cells. Each segment had a `min-width: 3px` floor that pushed the 60-cell bar past the column width and clipped the rightmost (today's) cell under `overflow: hidden`; segments now flex without a min-width.
Uptime percentage no longer reads `1.000%` for a 100%-up monitor. The `pct()` helper returned a `0..1` fraction while the UI rendered it as `%`; it now returns `0..100`.
Empty-state for new monitors: the list shows `awaiting first check…` instead of a row of misleading gray segments when no day has data yet.
Add Monitor modal no longer dismisses when a text drag-selection starts inside the dialog and the mouse is released over the backdrop. Closing now requires both mousedown and mouseup to land on the backdrop.

Apr 20, 2026

v0.4.2

Fix

v0.4.2

Monitors dashboard no longer freezes on first click. `MonitorsList` was running a no-op `useReactTable` (one dummy column, filtering done before the table) whose event-dispatch interaction with React 19 pinned the main thread on every click. Iterate the filtered rows directly and drop the `@tanstack/react-table` dependency — the table was never actually doing anything.

Apr 20, 2026

v0.4.1

Fix

v0.4.1

Dashboard no longer freezes after the first render in production. The deployed Next.js build was running with the experimental React Compiler enabled, which produced a tree where event handlers never wired up — the page rendered but every click was inert. The compiler is now disabled until the offending component is isolated.

Apr 20, 2026

v0.4.0

Feature

v0.4.0

Monitor dashboard now reads real check data. `GET /api/monitors/:id/stats` and `GET /api/monitors/stats/summary` expose per-monitor uptime (24h/7d/30d/90d), p50/p95/p99 latency, 90-day daily segments, a 120-sample response-time sparkline, the most recent checks, and recent incidents. The monitors list and detail pages use these instead of the previous placeholder numbers; a new monitor shows an explicit "waiting for first check" state until the probe records a sample.

Apr 20, 2026

v0.3.1

Fix

v0.3.1

Allow CHANGELOG.md through .dockerignore for web image build

Apr 20, 2026

v0.3.0

Feature

v0.3.0

Public roadmap at `/roadmap` with Now / Next / Later / Shipped buckets (T-179).
Placeholder docs pages for the upcoming Terraform provider (`/docs/terraform`, T-180) and CLI (`/docs/cli`, T-181), surfaced from the docs hub.
About page at `/about` covering mission, Teravor product positioning, and how to get in touch (T-183).
Web Docker image build now includes `CHANGELOG.md` so the `/changelog` page prerender succeeds.

Apr 20, 2026

v0.2.0

Feature

v0.2.0

Prorated upgrades and deferred downgrades for Whop billing, with a grace window and email reminders to uncancel or resubscribe (T-138).
Refreshed the Team & billing page with the new design handoff — plan card, usage meters, invoice list, upgrade CTA (T-137a).

Apr 19, 2026

v0.1.8

Fix

v0.1.8

Corrected the sandbox Whop API host to `sandbox-api.whop.com` so checkout sessions reach the right environment.

Apr 19, 2026

v0.1.7

Fix

v0.1.7

Routed the server-side Whop SDK at the sandbox API when `WHOP_ENVIRONMENT=sandbox` to stop cross-environment calls.

Apr 19, 2026

v0.1.6

Fix

v0.1.6

Surfaced the real Whop error and added a request timeout so Cloudflare 502s now carry a diagnosable cause instead of hanging.

Apr 19, 2026

v0.1.5

Fix

v0.1.5

Stopped the checkout-session request flood from retrying transient errors and started returning typed error codes to the client.

Apr 19, 2026

v0.1.4

Fix

v0.1.4

Moved the `/health` probe under `/api` so deploy healthchecks match production routing.
Made the Postfix probe reliable and gated Mailpit to dev-only compose.
Redirected `migrate` stdin from `/dev/null` during deploy so the step no longer hangs waiting for a TTY.
Force-recreated Caddy when `Caddyfile` changes so bind-mount inode updates take effect.
Unblocked `/docs/api` 500s and hardened the deploy script's failure paths.

Apr 19, 2026

v0.1.3

Feature

v0.1.3

`scripts/release.sh` — cuts a release by picking the next semver tag from Conventional Commits since the last tag, tagging, pushing, waiting for the GHCR build, and deploying to the Hetzner host.
Wired landing-page footer links to real destinations, gated the broken ones behind TODOs (T-187), and moved the Scalar API reference from `/docs` to `/docs/api` so the marketing docs hub could live at `/docs`.

Apr 19, 2026

v0.1.2

Fix

v0.1.2

Added the `config` dependency so `node-pg-migrate` runs under Bun.

Apr 19, 2026

v0.1.1

Fix

v0.1.1

Lazy-initialized the Postgres pool so the sitemap build no longer requires `DATABASE_URL` at build time.

Apr 19, 2026

v0.1.0

Feature

First tagged release. Closes Phase 0 and lands most of Phase 1 (MVP launch). The app can now run end-to-end: signup-by-waitlist → monitor → alert → pay.

Platform foundation: Bun + Hono API, Go probe worker, Next.js 16 dashboard, Postgres 18 + TimescaleDB + PGMQ + pg_cron (Phase 0 epics 0.A–0.H).
Scheduler with leader election via `pg_try_advisory_lock`, dispatch to PGMQ, structured tick metrics (T-040–T-044).
Probe worker: PGMQ consumer, HTTP checker with TLS capture and redirect policy, netguard that rejects private/reserved IPs, incident open/close in the same transaction (T-050–T-059).
Better Auth with the organization plugin, API-key and session auth middleware, monitors CRUD (T-030–T-039).
Additional monitor types: DNS, SSL, TCP, ICMP, Heartbeat/cron, plus keyword match for HTTP (T-100–T-106).
Alert engine that consumes the `alerts` queue with fire/suppress decisions, de-duplication, and cooldown windows (T-110, T-111).
Notifier fleet: email via self-hosted Postfix + OpenDKIM, Slack Block Kit, Discord embeds, generic webhook; exponential-backoff retry with dead-letter; first-notified timestamp on incidents (T-112, T-113, T-114, T-115, T-116, T-117).
Alert channel CRUD, per-monitor channel assignment, integrations page with test-fire and last-delivery (T-118, T-119, T-119a).
Public status pages at `/s/[slug]` with ISR, uptime percentages, five UI variants, admin editor, robots/sitemap (T-120–T-125a).
Whop billing: subscription upserts via signed webhooks, plan-tier entitlements, monitor-limit enforcement, overage surcharges, upgrade/downgrade flow with embedded checkout, dunning sweep (T-130–T-137).
Admin panel at `/admin`: org inspector, plan overrides, suspend/unsuspend, ops metrics, incidents view, abuse runbook, ownership verification for raw-IP monitors, per-target rate limits (T-140–T-146a).
API hardening: per-API-key token-bucket rate limits, scoped keys (read/write), HMAC-signed webhooks, OpenAPI reference at `/docs/api` via Scalar (T-150–T-153).
Marketing site: landing page, pricing calculator, competitor comparison pages, `/docs` hub, Getting Started guide, monitor-type reference, privacy/terms/DPA drafts, launch checklist, onboarding email template (T-160–T-168).
Invite-only access: waitlist submission, admin approval + invite-email flow, single-use token redemption, gated Better Auth registration (T-170–T-177).
Deployment: API/probe/web Dockerfiles with standalone output, Compose topology with Caddy, Cloudflare TLS via Let's Encrypt DNS-01, release/deploy scripts (Phase 0 Epic 0.H + T-075).
Dropped the app-owned `orgs` table; Better Auth's organization plugin now owns it natively.
Embedded Whop checkout in-app instead of redirecting externally.
React Compiler errors and a11y warnings surfaced by `react-doctor`.
Whop SDK routing through `@whop/api` and the correct `WHOP_API_BASE_URL` (v1, not v5).

Changelog

What's new.

Subscribe to the RSS feed or follow @teravor.

May 5, 2026

v0.10.0

Feature

v0.10.0

Dashboard status-page list view: stat cards (all/published/draft counts with plan-limit display), search by title or slug, published/draft filter bar, grid table with brand-color dot, slug, component count, status chip, relative updated time, View and Delete actions. Create button is plan-limit-gated (disabled with upgrade CTA when at limit). Removes the single-page UI guard that skipped the list when no pages existed (T-230).
Status page uptime percentage no longer double-scales the value (was multiplying by 100 a second time after the `pct()` helper already returned 0–100).

May 5, 2026

v0.9.4

Feature

v0.9.4

Focus trap on hand-built dialog components (Add Monitor, channel form, shortcuts, billing confirm) for keyboard accessibility.
`loading.tsx` skeleton fallbacks for app, marketing, admin, billing, and monitors route segments — no more blank screens during navigation.
ARIA tab roles on the monitor detail tab bar (`role="tablist"`, `role="tab"`, `aria-selected`, `aria-controls`, `role="tabpanel"`).
`aria-label` and `role="status"` on `StatusDot` so screen readers announce monitor state.
New migration: partial index on `check_results(monitor_id, checked_at DESC) WHERE status = 'down'` for heartbeat-sweep performance.
New migration: 90-day retention on `whop_webhook_events` via `pg_cron` + `prune_whop_webhook_events()`.
New migration: 30-day retention on dead-lettered `notification_attempts` via `pg_cron`.
New migration: index on `notification_attempts(incident_id, event, channel_id)` for delivery-summary GROUP BY.
Go probe: config validation tests, SSL and ICMP monitor config tests.
`BETTER_AUTH_SECRET` now rejects the known development default when `NODE_ENV=production`, preventing trivially forgeable session tokens.
`WHOP_PLANS_JSON` now warns when empty in production — an empty config previously caused every new subscription to silently fall back to the `'pro'` tier.
Go probe: DNS checker resolves the resolver hostname once and dials the IP directly, closing a DNS-rebinding TOCTOU that could bypass netguard on the second resolution.
Go probe: `processBatch` now caps concurrent goroutines with a semaphore (max 50), preventing resource exhaustion at large `BatchSize` values.
Go probe: dial timeout derived from per-check timeout (capped at 5 s) instead of hardcoded 10 s.
Go probe: `BatchSize` clamped to max 100, `PollInterval` floored at 50 ms, `PROBE_REGION` validated as alphanumeric on startup.
Go probe: HTTP URL validation rejects empty hosts and URLs with embedded credentials; TCP `send` capped at 10 KB; DNS `resolver` validated as `host:port`.
Go probe: ICMP echo ID uses `crypto/rand` instead of PID, reply buffer allocated once before the inner loop, body-match uses `bytes`/`[]byte` instead of `string` to halve peak memory.
Status page uptime stats now query the `check_results_daily` CAGG instead of the raw hypertable (~90 rows vs ~129 k per monitor).
Admin org detail endpoint runs 4 independent queries in `Promise.all` instead of sequentially.
Heartbeat sweep uses a single batch query for stale-monitor "recent down" checks instead of N+1 per-monitor queries.
Duplicated advisory-lock acquisition consolidated into a single `acquireAdvisoryLock()` in `lib/leader.ts`; notifiers import the shared version.
Duplicated `toIso` helper extracted to shared `lib/dates.ts` from three service files.
Shared `API_INTERNAL_URL` constant in `lib/config.ts` replaces three separate definitions.
`alert-channels` PATCH config check uses `!== undefined` instead of truthiness, fixing validation skip on `{}`.
IP rate-limiter and waitlist rate-limiter Maps now periodically evict stale entries (60 s interval, unref'd).
Invoice cache now periodically evicts expired entries.
Signup page Terms and Privacy links pointed at `/legal/terms` and `/legal/privacy` (404s); now correctly link to `/terms` and `/privacy`.
`sql.lit()` for user-supplied tag parameter replaced with parameterized `ARRAY[${tag}]::text[]`.
`sql.raw()` in dead-letter replay now guarded by an explicit allowlist check.
Invite redemption token restoration now runs in a `db.transaction()` on signup failure, preventing a TOCTOU broken-invite state.
Waitlist, admin-waitlist, and invite error responses now throw `HTTPException` consistently instead of returning ad-hoc JSON shapes (some missing `message` field).
Incident `scope` query parameter now validated through the Zod schema instead of raw cast.
`io.ReadAll` error in Go HTTP checker now produces a proper down result instead of a false body-match failure.
`downResult` in Go HTTP checker now guards against nil error to prevent a panic.
Go probe handlers map now protected by `sync.RWMutex` with a `started` guard to prevent data races if `Register` is called after `Run`.
Go probe `SetDeadline` errors now logged instead of silently discarded.
Empty catch blocks in ownership verification now log at debug level for diagnosability.
`console.error` in pool error handlers replaced with pino structured logger.
Grace-reminder processor now logs a warning when `downgrade_grace_until` is set but `pending_plan` is null (data inconsistency).
Swallowed `api_key.last_used_at` update error now logged with `logger.warn`.
`apiFetch` now throws `ApiError` when the response body contains an `{error}` key, catching mismatched error shapes at the boundary.
`add-monitor-flow` draft state typed as `MonitorDraft` instead of `any`.
`team-billing` `PLANS.find()` non-null assertions replaced with safe fallbacks.
`status-page-editor` `window.confirm()` replaced with the project's `AlertDialog` component.
Non-functional subscribe forms on status pages replaced with "coming soon" text.
Toast timeout now cleared on unmount to prevent `setState` on unmounted component.
Docs scroll spy throttled with `requestAnimationFrame` to avoid expensive per-pixel DOM reads.
Footer copyright year is now dynamic instead of hardcoded `2026`.
`.env.example` Whop sandbox keys replaced with placeholder values.

May 5, 2026

v0.9.3

Fix

v0.9.3

Caddy no longer marks the entire API upstream as unhealthy when a single route returns 5xx — passive health checks now only trigger on connection failures.

May 5, 2026

v0.9.2

Fix

v0.9.2

Zero-downtime deployments: Caddy passive health checks, API drain on shutdown, and warmup timing eliminate 521 errors during container rollover.
`PATCH /api/monitors/:id` config validation now validates against the specific monitor type's schema instead of the full union, preventing false rejections of valid edits.

May 4, 2026

v0.9.1

Fix

v0.9.1

API route validation schemas now use typed Zod unions (`httpConfigSchema`, `emailConfig`, `brandingSchema`, etc.) instead of `z.unknown()`, improving OpenAPI spec accuracy and request/response validation for monitors, alert channels, and status pages.

May 4, 2026

v0.9.0

Feature

v0.9.0

`/admin/ops` now shows a **Cron jobs · pg_cron** card listing the last 5 runs per scheduled job, including status, return message, start time, and duration. The new `GET /api/admin/ops/cron-runs` endpoint queries `cron.job_run_details` and emits a Sentry warning when the most recent run for any `argus-*` job has `status = 'failed'`.
`/admin/ops` queues card now shows a yellow `LAGGY` chip when `oldest_msg_age_sec > 60` and a red `STUCK` chip when `> 300`, so queue health is visible at a glance.
API emits a Sentry `warning` when a PGMQ queue crosses the LAGGY or STUCK threshold, deduped per-queue per-15-minutes to avoid alert fatigue.
**Sentry release tagging:** `SENTRY_RELEASE` env var is now wired into the API and probe Sentry init blocks. `scripts/deploy.sh` passes `TAG` as `SENTRY_RELEASE` so every deployed image is tagged with its commit SHA in Sentry, enabling accurate "issue introduced in" timelines and per-release regression tracking.
Security page at `/security` with accurate infrastructure details: Hetzner Falkenstein hosting, Caddy TLS, bcrypt/Better Auth, SSH key-only access, and responsible disclosure contact. Footer link re-enabled.
Bulk monitor operations: select multiple monitors on the dashboard with per-row checkboxes and a "select all" header checkbox; a floating action bar offers Pause, Resume, and Delete (with confirmation dialog) for all selected monitors at once.
API endpoints `POST /api/monitors/bulk/pause`, `/bulk/resume`, and `/bulk/delete` — cap at 100 IDs per request, scope-checked against the requesting org, return `{ affected: number }`.
Editing a monitor in the Settings tab no longer gets wiped by background 30-second refetches. The form now tracks a `dirty` flag and skips the prop-to-state sync while the user has unsaved changes; syncing resumes after a successful save.
Replaced array-index React keys with stable values in the monitor assertions Chip lists (`monitor-assertions.tsx`) and the plan feature list in the billing page (`team-billing.tsx`), preventing spurious remounts when lists update.
Dashboard topbar: removed non-functional search input, "Public status ↗" placeholder link, and notifications bell (all were decorative with no implementation).

May 1, 2026

v0.8.2

Fix

v0.8.2

The 60-day/90-day uptime bar on the dashboard rendered the rightmost ~1 day as `unknown` (gray) even when probe data was up-to-date. The `check_results_daily` continuous aggregate added in v0.8.0 inherited TimescaleDB's post-2.13 default of `materialized_only = true`, so reads saw only the materialized portion — and the refresh policy can't materialize today's bucket (the bucket only completes after day boundary + the 1h `end_offset`). New migration enables realtime mode (`materialized_only = false`); reads now union the materialized rollup with a real-time computation against the raw hypertable for any bucket past the watermark.

May 1, 2026

v0.8.1

Fix

v0.8.1

Deploy migration `1776988800046_check_results_indexes.sql` failed against production because Timescale hypertables don't support `CREATE INDEX CONCURRENTLY` (the planner propagates the index to all chunks in one transaction; node-pg-migrate also runs each migration inside a transaction). Switched to a plain `CREATE INDEX IF NOT EXISTS` so the next deploy applies cleanly. The v0.8.0 deploy job that errored on this migration left no `pgmigrations` row, so re-running picks up where it stopped.

Apr 30, 2026

v0.8.0

Feature

v0.8.0

Sentry distributed tracing on API HTTP requests and probe operations, sampled via the new `SENTRY_TRACES_SAMPLE_RATE` env var (`0` disables; `0.1` is the prod default in the docs). API transactions are named after the matched Hono route pattern so `/api/monitors/:id` doesn't explode into one transaction per id. Slow endpoints now show up as span chains in Sentry Performance instead of being invisible.
Tick errors in every batch processor (scheduler, alert engine, four notifiers, overage processor, grace-reminder processor) now reach Sentry with a `module` tag, instead of being logged and dropped. The shared runner captures once at the catch-all so the per-processor code stays clean; the scheduler's heartbeat-write swallow gets the same treatment so a sustained outage of the ops-metrics row shows up in Sentry alongside the STALE chip on `/admin/ops`.
Explicit Postgres connection-pool sizing on the API (`DB_POOL_MAX`, `DB_POOL_MIN`, `DB_POOL_IDLE_MS`, `DB_POOL_CONNECTION_TIMEOUT_MS`). Defaults to a conservative 10-connection cap that fits a single API replica plus the scheduler, four notifiers, and Better Auth pool against a small Postgres instance. Pools no longer crash the process on a transient idle-client drop.
TimescaleDB compression and retention policies on `check_results`: chunks older than 7 days are compressed (segmented by `monitor_id`, ordered by `checked_at DESC`) and chunks older than 100 days are dropped. The dashboard hits a 90-day window; 10 days of buffer protects boundary skew. Storage growth on the time-series table is now bounded.
Daily retention sweep on `notification_attempts` via `pg_cron`: the new `prune_notification_attempts(retain_days)` function deletes rows older than 60 days every night at 03:30 UTC. The Alerts tab on monitor detail joins this table for delivery status, but the dashboard's incident window only reaches back 30 days, so 60-day retention keeps every queryable incident's delivery trail intact while bounding table growth.
TimescaleDB continuous aggregate `check_results_daily` rolls up per-monitor per-day check counts (`n`, `up_count`, `bad_count`). Refresh policy slides the last 7 days forward every 30 minutes with a 1-hour end-offset; realtime mode keeps the un-materialized last hour fresh against the raw table. Sets up the foundation for the next stats-query rewrite to read aggregates instead of scanning raw time-series.
Admin **Dead letters** panel on `/admin/ops` — surfaces every dead-lettered notification (retry-exhaustion or channel-type mismatch) with queue, monitor, attempt count, last error, and archived time. New `GET /api/admin/ops/dead-letters` endpoint backs it. Operators no longer have to query PGMQ archive tables by hand to find what failed.
**Replay** button on each Dead-letters row (`POST /api/admin/ops/dead-letters/:id/replay`) re-enqueues the original message onto its source queue with the attempt counter reset, leaves a `replayed` audit row in `notification_attempts`, and is idempotent under double-clicks. Returns 410 when the archived payload has aged out and 409 when it has already been replayed.
Improved mobile responsiveness across landing page and app dashboard: collapsible sidebar drawer with backdrop overlay (hamburger button in topbar), responsive landing nav with mobile hamburger menu, all landing section grids adapt from single column on phones through desktop multi-column layouts, section padding scales with `clamp()`, monitors table wraps in a horizontal-scroll container, 90-day uptime bar scrolls horizontally instead of overflowing, stat card grids collapse to 2 columns on small screens, Add Monitor modal caps at 92 vw on narrow viewports (T-216).
Standardized empty- and error-state surfaces across the dashboard. New shared `<EmptyCard>` and `<ErrorCard>` primitives replace ad-hoc inline strings on the monitor detail (load + stats), response log, alerts tab, monitors list, and alert-channels list. Errors are sanitized: `ApiError` 401/403/404/429/5xx and network failures get user-friendly copy, with the raw message tucked inside a collapsible "Technical details" block instead of dumped inline.
Dashboard detail-page polling now pauses while the tab is hidden. The four monitor query hooks (`useMonitorStatsQuery`, `useMonitorChecksQuery`, `useMonitorAlertEventsQuery`, and the list-view `useMonitorsStatsQuery`) now set `refetchIntervalInBackground: false`; refetches resume the moment the tab regains focus.
Monitor dashboard's 90-day uptime bar (and the list view's 60-day variant) now read from the `check_results_daily` continuous aggregate instead of scanning raw `check_results`. The dashboard pulls ~90 pre-aggregated rows per monitor on each poll instead of grouping over ~129k raw rows for a 60s-interval monitor. Sub-day-granularity queries (latest checks, sparkline, 24h uptime/latency) still read the raw table because they only touch one or two recent chunks.
Tagged releases now deploy automatically. Pushing a `v*` tag (typically via `scripts/release.sh`) triggers `release.yml`, which builds the three images, pushes them to GHCR, and then SSHes into the Hetzner host to run the existing `scripts/deploy.sh`. The laptop no longer needs SSH access for routine deploys; `scripts/release.sh` just bumps the version, promotes the CHANGELOG, pushes the tag, and watches CI. Manual `scripts/deploy.sh` remains supported for emergencies.
Alert engine: a `closed` alert against an incident that is still open is now suppressed with reason `incident_still_open` instead of being fired. The previous behavior was asymmetric — `opened` against an already-closed incident was guarded, but `closed` against a still-open incident was not — which meant a stale or replayed `closed` payload could trigger a recovery email on a live outage.
Probe consumer: a check that fails with the same error each retry no longer thrashes the queue at the static visibility-timeout cadence. The consumer now pushes each retry past an exponential backoff (`PROBE_BACKOFF_BASE_SEC`, default 30s, doubling to `PROBE_BACKOFF_MAX_SEC`, default 600s, ±20% jitter) and archives the message proactively on the final attempt instead of burning one more VT cycle just to detect "still failing".
Heartbeat scheduler sweep: dropped the `COALESCE(last_ping_at, created_at)` fallback that compensated for never-pinged heartbeat monitors. Going forward, `createMonitor` seeds `last_ping_at` to creation time and a CHECK constraint enforces non-null on heartbeat type, so the sweep query and its index can run on a single column. Behaviour is unchanged from the operator's perspective — a heartbeat monitor opens an incident at exactly `interval + grace` after the last (or seeded) ping.
Notifiers: an alert routed to a channel of the wrong type (e.g. an email payload pointing at a Slack channel id, possible only on schema drift or alert-engine fanout bugs) is now dead-lettered with a typed `notification_attempts` row and a Sentry exception. Previously such payloads were silently archived with no record, hiding the fanout bug behind apparent silence.
CI: alert-engine processor tests and the Go probe round-trip seeder now set `last_ping_at` when inserting heartbeat monitors directly, so they satisfy the `monitors_heartbeat_last_ping_at_check` CHECK constraint added with the heartbeat-sweep cleanup. (Previously both bypassed the service-layer `createMonitor()` helper that already seeded the column, and the constraint refused the insert at test time.)

Apr 21, 2026

v0.7.0

Feature

v0.7.0

Self-hosted mailbox on `mail.teravor.com`: receive at role addresses (`admin@`, `alerts@`, `abuse@`, `postmaster@`, `hello@`, `support@`, `contact@`) and a catch-all — all delivering to one IMAP inbox. Powered by Stalwart Mail Server (replaces Postfix).
Contact page at `/contact` with support and security-report email addresses.

Apr 21, 2026

v0.6.0

Feature

v0.6.0

Interval floor enforcement in the add-monitor and monitor-settings forms: interval buttons below the plan minimum are disabled with upgrade tooltips, and the add-monitor flow defaults to the fastest interval the current plan allows.
Monitor count gate in the add-monitor flow: a banner on step 3 shows the limit and an upgrade link when the org is at capacity; the Create button is disabled until the limit is lifted.
Sidebar plan label, nav counts (monitors, status pages, integrations), and the monitors-list stat card are now driven by live API data instead of hardcoded values.
Hardcoded `monitor.teravor.com` domain in the status-page editor, status embed widget, and public status page route replaced with `NEXT_PUBLIC_APP_URL`.

Apr 21, 2026

v0.5.1

Fix

v0.5.1

Production email delivery: the Postfix relay advertised STARTTLS without a server cert, so nodemailer auto-negotiated TLS and every send failed with `454 4.7.0 TLS not available due to local problem` (all notifications dead-lettered). Switched the internal :587 relay to `smtpd_tls_security_level=none` since traffic stays on the Docker bridge; the outbound hop to remote MXes still uses opportunistic TLS. Also swapped the flaky `nc`-based healthcheck for `postfix status`.
Postfix SASL credentials were silently empty in production, causing `454 4.7.0 Temporary authentication failure: generic failure` on every notifier send. Two layered bugs: compose-time `${VAR}` substitution reads `.env` from the compose file's directory (`docker/`), not the repo root, so `${POSTFIX_SMTP_USER:-}` expanded to `""`; and boky/postfix provisions its SASL account from `SMTPD_SASL_USERS` (format `user:pass`), not `SMTP_USERNAME`/`SMTP_PASSWORD` (those are for outbound *relay* auth, unused here). The service now uses `env_file: ../.env` plus `POSTFIX_smtpd_sasl_local_domain=teravor.com` so nodemailer's bare-username SASL login matches the `[email protected]` entry in sasldb. `.env.example` and `docs/email-notifier.md` updated.

Apr 20, 2026

v0.5.0

Feature

v0.5.0

Monitor detail page: the **Response Log**, **Assertions**, and **Alerts** tabs are now implemented (previously stubs). Response Log paginates through raw check results newest-first. Assertions renders a read-only view of the monitor's configured assertions (expected status codes, body match, DNS records, SSL expiry thresholds, TCP expect/send, heartbeat grace) alongside a recent-failures list. Alerts shows assigned delivery channels plus an alert-engine decision history with per-channel delivery status.
API: `GET /api/monitors/:id/checks` and `GET /api/monitors/:id/alert-events` — paginated (ISO-timestamp `before` cursor) endpoints backing the new tabs.
Monitor **Settings** tab is now a full edit surface: every type-specific config field (URL, method, expected status, body match, DNS record/values, SSL warn/critical days, TCP host/port/send/expect, ICMP count/packet size, heartbeat grace) plus check interval, enabled state, friendly name, and alert channels can be edited in place. Reuses the same type-specific forms as the Add Monitor flow.
Monitor detail Overview: the disabled "Full response log →" button now navigates to the Response Log tab, and the "Assertions" placeholder card becomes a nav card to the Assertions and Alerts tabs.
API: `PATCH /api/monitors/:id` now accepts a typed `config` body (validated per monitor type via the same Zod schemas as create). The monitor `type` remains immutable — changing type still requires delete + create.
API: when a config edit moves a raw-IP monitor to a new host, the monitor is disabled and the response carries a fresh ownership challenge (`token`, `well_known_url`, `dns_name`); the caller must re-run `POST /api/monitors/:id/verify-ownership` before probes resume.

Apr 20, 2026

v0.4.3

Feature

v0.4.3

Monitor detail "Edit" button is now functional — jumps to the Settings tab, which gained an editable Friendly name field. Save persists name and channel changes together, and works even when no alert channels are configured.
Add Monitor flow auto-derives a friendly name from the monitor config (HTTP hostname, `host` for DNS/TCP/SSL/ICMP, dated label for heartbeats) when the user leaves the field blank, instead of the monitor surfacing as a UUID slice.
Monitors list falls back to the config `host` (not the id slice) when a monitor has no name, so DNS/TCP/SSL/ICMP monitors display sensibly without a name.
Monitors list 60-day uptime bar now actually shows the colored cells. Each segment had a `min-width: 3px` floor that pushed the 60-cell bar past the column width and clipped the rightmost (today's) cell under `overflow: hidden`; segments now flex without a min-width.
Uptime percentage no longer reads `1.000%` for a 100%-up monitor. The `pct()` helper returned a `0..1` fraction while the UI rendered it as `%`; it now returns `0..100`.
Empty-state for new monitors: the list shows `awaiting first check…` instead of a row of misleading gray segments when no day has data yet.
Add Monitor modal no longer dismisses when a text drag-selection starts inside the dialog and the mouse is released over the backdrop. Closing now requires both mousedown and mouseup to land on the backdrop.

Apr 20, 2026

v0.4.2

Fix

v0.4.2

Monitors dashboard no longer freezes on first click. `MonitorsList` was running a no-op `useReactTable` (one dummy column, filtering done before the table) whose event-dispatch interaction with React 19 pinned the main thread on every click. Iterate the filtered rows directly and drop the `@tanstack/react-table` dependency — the table was never actually doing anything.

Apr 20, 2026

v0.4.1

Fix

v0.4.1

Dashboard no longer freezes after the first render in production. The deployed Next.js build was running with the experimental React Compiler enabled, which produced a tree where event handlers never wired up — the page rendered but every click was inert. The compiler is now disabled until the offending component is isolated.

Apr 20, 2026

v0.4.0

Feature

v0.4.0

Monitor dashboard now reads real check data. `GET /api/monitors/:id/stats` and `GET /api/monitors/stats/summary` expose per-monitor uptime (24h/7d/30d/90d), p50/p95/p99 latency, 90-day daily segments, a 120-sample response-time sparkline, the most recent checks, and recent incidents. The monitors list and detail pages use these instead of the previous placeholder numbers; a new monitor shows an explicit "waiting for first check" state until the probe records a sample.

Apr 20, 2026

v0.3.1

Fix

v0.3.1

Allow CHANGELOG.md through .dockerignore for web image build

Apr 20, 2026

v0.3.0

Feature

v0.3.0

Public roadmap at `/roadmap` with Now / Next / Later / Shipped buckets (T-179).
Placeholder docs pages for the upcoming Terraform provider (`/docs/terraform`, T-180) and CLI (`/docs/cli`, T-181), surfaced from the docs hub.
About page at `/about` covering mission, Teravor product positioning, and how to get in touch (T-183).
Web Docker image build now includes `CHANGELOG.md` so the `/changelog` page prerender succeeds.

Apr 20, 2026

v0.2.0

Feature

v0.2.0

Prorated upgrades and deferred downgrades for Whop billing, with a grace window and email reminders to uncancel or resubscribe (T-138).
Refreshed the Team & billing page with the new design handoff — plan card, usage meters, invoice list, upgrade CTA (T-137a).

Apr 19, 2026

v0.1.8

Fix

v0.1.8

Corrected the sandbox Whop API host to `sandbox-api.whop.com` so checkout sessions reach the right environment.

Apr 19, 2026

v0.1.7

Fix

v0.1.7

Routed the server-side Whop SDK at the sandbox API when `WHOP_ENVIRONMENT=sandbox` to stop cross-environment calls.

Apr 19, 2026

v0.1.6

Fix

v0.1.6

Surfaced the real Whop error and added a request timeout so Cloudflare 502s now carry a diagnosable cause instead of hanging.

Apr 19, 2026

v0.1.5

Fix

v0.1.5

Stopped the checkout-session request flood from retrying transient errors and started returning typed error codes to the client.

Apr 19, 2026

v0.1.4

Fix

v0.1.4

Moved the `/health` probe under `/api` so deploy healthchecks match production routing.
Made the Postfix probe reliable and gated Mailpit to dev-only compose.
Redirected `migrate` stdin from `/dev/null` during deploy so the step no longer hangs waiting for a TTY.
Force-recreated Caddy when `Caddyfile` changes so bind-mount inode updates take effect.
Unblocked `/docs/api` 500s and hardened the deploy script's failure paths.

Apr 19, 2026

v0.1.3

Feature

v0.1.3

`scripts/release.sh` — cuts a release by picking the next semver tag from Conventional Commits since the last tag, tagging, pushing, waiting for the GHCR build, and deploying to the Hetzner host.
Wired landing-page footer links to real destinations, gated the broken ones behind TODOs (T-187), and moved the Scalar API reference from `/docs` to `/docs/api` so the marketing docs hub could live at `/docs`.

Apr 19, 2026

v0.1.2

Fix

v0.1.2

Added the `config` dependency so `node-pg-migrate` runs under Bun.

Apr 19, 2026

v0.1.1

Fix

v0.1.1

Lazy-initialized the Postgres pool so the sitemap build no longer requires `DATABASE_URL` at build time.

Apr 19, 2026

v0.1.0

Feature

First tagged release. Closes Phase 0 and lands most of Phase 1 (MVP launch). The app can now run end-to-end: signup-by-waitlist → monitor → alert → pay.

Platform foundation: Bun + Hono API, Go probe worker, Next.js 16 dashboard, Postgres 18 + TimescaleDB + PGMQ + pg_cron (Phase 0 epics 0.A–0.H).
Scheduler with leader election via `pg_try_advisory_lock`, dispatch to PGMQ, structured tick metrics (T-040–T-044).
Probe worker: PGMQ consumer, HTTP checker with TLS capture and redirect policy, netguard that rejects private/reserved IPs, incident open/close in the same transaction (T-050–T-059).
Better Auth with the organization plugin, API-key and session auth middleware, monitors CRUD (T-030–T-039).
Additional monitor types: DNS, SSL, TCP, ICMP, Heartbeat/cron, plus keyword match for HTTP (T-100–T-106).
Alert engine that consumes the `alerts` queue with fire/suppress decisions, de-duplication, and cooldown windows (T-110, T-111).
Notifier fleet: email via self-hosted Postfix + OpenDKIM, Slack Block Kit, Discord embeds, generic webhook; exponential-backoff retry with dead-letter; first-notified timestamp on incidents (T-112, T-113, T-114, T-115, T-116, T-117).
Alert channel CRUD, per-monitor channel assignment, integrations page with test-fire and last-delivery (T-118, T-119, T-119a).
Public status pages at `/s/[slug]` with ISR, uptime percentages, five UI variants, admin editor, robots/sitemap (T-120–T-125a).
Whop billing: subscription upserts via signed webhooks, plan-tier entitlements, monitor-limit enforcement, overage surcharges, upgrade/downgrade flow with embedded checkout, dunning sweep (T-130–T-137).
Admin panel at `/admin`: org inspector, plan overrides, suspend/unsuspend, ops metrics, incidents view, abuse runbook, ownership verification for raw-IP monitors, per-target rate limits (T-140–T-146a).
API hardening: per-API-key token-bucket rate limits, scoped keys (read/write), HMAC-signed webhooks, OpenAPI reference at `/docs/api` via Scalar (T-150–T-153).
Marketing site: landing page, pricing calculator, competitor comparison pages, `/docs` hub, Getting Started guide, monitor-type reference, privacy/terms/DPA drafts, launch checklist, onboarding email template (T-160–T-168).
Invite-only access: waitlist submission, admin approval + invite-email flow, single-use token redemption, gated Better Auth registration (T-170–T-177).
Deployment: API/probe/web Dockerfiles with standalone output, Compose topology with Caddy, Cloudflare TLS via Let's Encrypt DNS-01, release/deploy scripts (Phase 0 Epic 0.H + T-075).
Dropped the app-owned `orgs` table; Better Auth's organization plugin now owns it natively.
Embedded Whop checkout in-app instead of redirecting externally.
React Compiler errors and a11y warnings surfaced by `react-doctor`.
Whop SDK routing through `@whop/api` and the correct `WHOP_API_BASE_URL` (v1, not v5).