Changelog

Improved

• Graceful error & loading states — an unexpected error while loading a dashboard page now shows a recoverable 'try again' screen instead of a blank or broken page, and pages show a lightweight skeleton while they load. Unmatched URLs get a proper on-brand 404.
• No more stale overview data — switching vendors, time ranges, or endpoints quickly now cancels the previous request, so a slower earlier response can't resolve late and overwrite the chart with the wrong vendor's numbers.
• Cleaner not-found handling — requests for malformed resource ids (API keys, notification channels, SLOs, maintenance windows) return a 404 instead of a 500.

Security

• Outbound URL hardening — webhook and Slack alert destinations are now resolved and validated before delivery, rejecting any that point at private, loopback, link-local, or cloud-metadata addresses. This closes a server-side request forgery (SSRF) vector where a crafted destination could be used to probe internal services. Redirects are no longer followed on alert delivery — each hop is re-validated, so a public URL can't bounce a request onto an internal host.
• Credential masking — Slack webhook URLs and PagerDuty integration keys are now masked (alongside webhook signing secrets) whenever a notification channel is read back, so credentials never leave the server in plaintext. Editing a channel without re-entering a credential now keeps the stored value intact instead of overwriting it.
• Signup key handling — the API key issued during signup is now held server-side only while you complete onboarding, instead of in your browser session. The one-time reveal on the onboarding screen is unchanged, and the key still expires after an hour if onboarding isn't completed.

Added

• SLA Reporting page (/dashboard/sla) — define p50/p95/p99 latency SLOs per vendor or endpoint; window toggle (30/60/90 days); calendar heatmap with met/missed/no-data coloring; compliance %, breaching hours, and longest breach run per SLO. Solo and Team plans.
• Vendor-wide SLOs — when no endpoint is selected, compliance is computed from a true vendor-level percentile (not an average of per-endpoint p95s, which is mathematically incorrect). A new latency_vendor_hourly materialized view handles this.
• Sample guard — buckets with fewer than 20 requests are excluded from compliance calculations to avoid noise from low-traffic periods.

Collector

• latency_vendor_hourly materialized view (migration 028) — vendor-level latency aggregates without the endpoint dimension; refreshed alongside latency_hourly by the existing refresh_aggregates job.
• slo_definitions table (migration 029) — stores per-customer SLO config (vendor, endpoint, metric, threshold, env, default window).
• GET /v1/slo, POST /v1/slo, DELETE /v1/slo/{id} — SLO definition CRUD (Solo/Team only).
• GET /v1/sla — compliance report: per-SLO compliance %, breach counts, longest breach run, and day-by-day heatmap data for the selected window.

Added

• Model Radar page (/dashboard/models) — table of every AI model your code has called, with Active / Deprecated / Sunset status badges and sunset dates. Available on all tiers.
• Deprecation alerts — daily job checks which deprecated or sunset models you called in the last 24 hours and fires a fan-out alert if you haven't been notified in the last 7 days.
• SDK v0.4.0 (Ruby, Python, JS) — extracts the model field from AI vendor JSON response bodies (OpenAI, Anthropic, Gemini, Mistral, Cohere) without consuming the response stream. Opt-out via capture_model_names: false.

Collector

• model_usage table (migration 027) — upserted on every event batch that carries a model name; tracks call_count, first_seen_at, and last_seen_at per customer/vendor/model/endpoint.
• GET /v1/model-usage — returns model list enriched with deprecation status from the in-repo data/deprecated_models.json catalog.
• model_deprecation_alert_history table (migration 027) — 7-day per-model deduplication for deprecation alerts.

Added

• apidepth setup — interactive CLI wizard in every SDK (bundle exec apidepth setup, python -m apidepth setup, npx apidepth setup). Detects your framework, generates the correct initializer snippet, and optionally writes it to disk. Fully non-interactive via --api-key + --no-prompt for CI/CD and AI-assisted setup.
• apidepth test — send a synthetic test event and confirm the collector received it, with per-failure-mode error messages (bad key, unreachable, SSL, timeout) and a hard 5 s timeout.
• Smart ignored-host defaults — localhost, 127.0.0.1, 0.0.0.0, ::1, and the collector URL itself are automatically ignored by all three SDKs. Glob patterns (*.internal, *.svc.cluster.local) are now supported everywhere.
• Framework-aware setup output — setup detects Rails, Sinatra, Django, FastAPI, Express, and Next.js from project files and generates the framework-correct initializer snippet.
• No-events-received email — if your account has no events 24 hours after signup, you'll receive a helpful re-engagement email. Suppressed when any event (including test events) has been received. Re-fires after an SDK version bump with no subsequent events.
• SDK version nudge — the overview page now surfaces per-SDK upgrade banners when your running version is behind the current stable. Breaking-change gaps show a prominent red notice with a migration guide link; minor gaps show a subdued amber banner. Dismissible per-account per version gap.

Collector

• New test_events table — test events are routed separately and never appear in the vendor list, latency charts, or alerts.
• New sdk_versions and dismissed_sdk_nudges tables — track current stable releases and per-account dismissals.
• Retention job now prunes test_events (30-day window) and activation_emails_sent (90-day window).

Added

• Environment switcher — a sidebar dropdown filters all dashboard data by environment tag (production, staging, or any value your SDK config passes). Burn-down alerts remain production-only.
• 1 h and 90 d time ranges — added to the overview and rate-limits pages alongside the existing 24 h, 7 d, and 30 d options.

Improved

• Vendor display names — AWS, GitHub, OpenAI, HubSpot, PagerDuty, and SendGrid now show correct casing across all pages.
• Active navigation — sidebar highlights the current page via client-side route detection; Settings and Notifications are treated as distinct entries.
• Empty states — latency, error rate, and rate-limit charts now show a quickstart nudge when no data has been captured yet.
• Vendor tab selection — active tab uses a cyan inset ring instead of a heavier background fill.
• Settings page — Pin, Hide, and Remove controls restyled as bordered ghost buttons; prose instructions replaced with a color-dot legend; Reset to defaults requires a two-step inline confirmation.

Added

• Maintenance windows — create time-bounded maintenance windows from the dashboard to suppress all alert fan-out during planned downtime. Solo and Team plans. Windows expire automatically; active windows are shown in the notifications settings page.
• GET /v1/environments — new collector endpoint returns the distinct environment tags seen for a customer, powering the environment switcher dropdown.

Fixed

• API key race condition (DASH-024) — SELECT FOR UPDATE added to key-creation path to prevent duplicate keys under concurrent requests.
• Missing collector URL (DASH-025) — dashboard now throws a clear startup error when NEXT_PUBLIC_COLLECTOR_URL is absent instead of silently failing.
• Vendor-config 404 (DASH-026) — Prisma P2025 (record not found) on custom vendor DELETE now returns 404 instead of 500.
• Slack webhook validation (DASH-027) — Slack channel test route checks for a webhook_url before attempting delivery.
• Free-tier vendor cap in transaction (COL-022) — cap enforcement is now atomic; previously a race could allow a 4th vendor through.
• Async email delivery (COL-023) — alert emails are sent in a background task so a slow Resend response no longer delays the collector request.
• Burn-down ratio inversion (COL-024) — rl_remaining / rl_limit was inverted; burn-down risk levels are now correct.
• rl_reset_at validation (COL-025) — malformed reset timestamps are discarded rather than written as epoch 0.
• Digest misfire grace period (COL-026) — weekly digest now waits a 5-minute grace window to avoid double-send during rolling deploys.
• Shared httpx client (COL-030) — probe and collector share a single httpx.AsyncClient with proper connection pooling; fixes resource exhaustion under load.

SDK

• Ruby v0.3.0 — gem 'apidepth', '~> 0.3'. SDK initialisation is now fully non-blocking: the registry fetch and background flush thread start in a daemon thread so slow collector responses never delay app boot. ReDoS protection added to the path-normalisation regex. Header parsing uses Net::HTTPResponse#[] (case-insensitive) instead of raw @header access.
• Python v0.2.2 — pip install apidepth==0.2.2. Cold-start detection via a per-process host registry — first request to each vendor in a process lifetime is flagged cold_start=True. All rate-limit header lookups now normalise to lowercase before extraction, fixing missed headers on servers that capitalise header names.

Added

• JavaScript / TypeScript SDK — npm install apidepth. Compatible with Node.js 18+.
• Auto-instrumentation — call Apidepth.configure({ apiKey }) then Apidepth.instrument() once at startup. Patches node:http, node:https, and globalThis.fetch — no per-request changes needed.
• Express middleware — app.use(apidepthMiddleware({ apiKey: process.env.APIDEPTH_API_KEY })) imported from apidepth/integrations/express.
• Next.js integration — register() in instrumentation.ts via apidepth/integrations/nextjs for zero-config App Router setup.
• Rate limit header extraction and cold-start detection included.

SDK

• npm install apidepth — first npm publish. Node.js 18+. Integrations for Express (apidepth/integrations/express) and Next.js (apidepth/integrations/nextjs) included.

Added

• Permanent free tier — the trial countdown is removed. Free accounts track up to 3 vendors indefinitely with 7-day data retention and email alerts.
• Natural upgrade prompt — the moment to upgrade is when your stack grows past 3 vendors, not a deadline. Solo plan adds unlimited vendors and 90-day retention.

Added

• Vendor Status panel — the overview dashboard now shows a live status pill for each API you depend on: Operational, Degraded, Outage, or Unreachable.
• Synthetic probes — probes run every 60 seconds across major vendors, independent of your SDK traffic.
• TLS certificate warnings — expiry notices shown inline when a certificate has fewer than 14 days remaining.
• Active incident summaries — detected incidents listed below each vendor's status pill.
• Works on quiet days — probe results are independent of SDK event volume, so vendor status is always current even if no requests are flowing.

Added

• Notification channels — configure multiple alert destinations per account at /settings/notifications. Each channel can filter by alert type (p95 threshold, error rate, rate limit, quota burn-down) and minimum severity.
• Slack — paste a Slack incoming webhook URL to receive rich formatted alert messages with vendor name, metric vs baseline, environment, and a dashboard link.
• Outbound webhooks — POST alert events as JSON to any HTTPS endpoint. Optional HMAC-SHA256 signing via X-Apidepth-Signature header (same convention as GitHub webhooks) so you can verify authenticity.
• Test delivery — every channel has a Test button that sends a live test payload and shows success or failure inline.
• Per-channel alert filtering — choose which alert types each channel receives and set a minimum severity (warning or critical only).
• Tier gating — Slack and webhook channels require Solo plan; PagerDuty (coming) requires Team. Trial accounts receive email alerts only.
• Python SDK — pip install apidepth. Mirrors the Ruby SDK feature set: passive requests and httpx instrumentation, vendor normalisation, rate limit header capture, and custom vendor support.

SDK

• pip install apidepth — first PyPI publish. Works with requests, httpx, and any library that wraps them.

Added

• Attribution card on the overview dashboard — surfaces a vendor-vs-app blame split for each vendor using real fleet data. Three verdicts: Isolated (your p95 is >20% above fleet median — investigate your code), Tracking (within ±20% of fleet — consistent with vendor-wide issue), Healthy (>20% below fleet — you're doing better than average).
• Fleet-grounded verdicts — requires at least 10 fleet contributors before showing a verdict; otherwise shows "insufficient fleet data" rather than a false signal.
• Honest copy — verdicts use "this pattern suggests" language, not "it is definitely Stripe." The goal is fast triage, not false certainty.

Added

• Weekly digest email — sent every Monday at 09:00 UTC to all active accounts. Covers per-vendor p95 trend vs the prior week, error rate summary, and 429 pattern highlights.
• Smart hook sentence — the email opens with the most important insight first: worst degradation, most improved vendor, or a quiet-week confirmation. No filler copy.
• Deduplication — a digest_history table prevents re-sends if the job runs twice (e.g. during deploys).

Added

• /status page — live uptime dashboard showing current status, 30d/90d uptime %, a 90-day day-bar history chart, and an incident list.
• Self-hosted health monitoring — the collector pings its own database every minute and records the result. No Betterstack, no third-party dependency — fully on-brand.
• GET /v1/status public collector endpoint (no auth required); powers the status page with current status, uptime percentages, daily breakdown, and incident history.
• Docs site complete — Getting Started, Configuration, Rate Limit Intelligence, and API Reference all live at /docs.
• Changelog page at /changelog.

Added

• Rate Limit Intelligence — tracks 429 responses hourly per vendor and endpoint; visible in the new Rate Limits dashboard page.
• Burn-down prediction — the Ruby SDK (v0.2.0) now extracts x-ratelimit-*, ratelimit-*, and retry-after headers from every response and sends rl_remaining, rl_limit, and rl_reset_at to the collector.
• Quota snapshots — one live row per customer/vendor/environment updated on every event batch; no storage bloat.
• Burn-down alerts — email notification when projected time to quota exhaustion drops below 20 minutes; 30-minute per-vendor cooldown.
• Reset format normalisation — three formats unified to epoch milliseconds: Unix timestamps, seconds-from-now, and OpenAI duration strings (1m30s, 20ms, 2h).
• Docs site — /docs with Getting Started, Configuration reference, Rate Limit Intelligence, and API reference.
• RubyGems badge on the landing page.

SDK

• gem 'apidepth', '~> 0.2' — publish gem v0.2.0 to RubyGems.

Added

• Passive instrumentation — Module#prepend on Net::HTTP captures every outgoing request with zero config changes to call sites. Works with Faraday, HTTParty, RestClient, and plain Net::HTTP.
• Vendor normalisation — built-in recognition for Stripe, OpenAI, Anthropic, Twilio, Resend, and GitHub; IDs and UUIDs stripped from paths automatically.
• Latency percentiles — p50/p95/p99 per vendor/endpoint/hour, displayed on the Overview dashboard.
• Fleet benchmarks — anonymised, opt-in fleet medians and p95s; your numbers plotted against real production data from all users (minimum 10 contributors per vendor/endpoint).
• Error rate tracking — client (4xx) and server (5xx) error rates per vendor/endpoint/hour.
• Timeout capture — Net::ReadTimeout and Net::OpenTimeout are recorded and re-raised; previously invisible in standard monitoring.
• Cold-start tagging — first request to a vendor is flagged so warmup latency doesn't inflate percentiles.
• Threshold alerts — email when your p95 exceeds 2× the 7-day baseline for a vendor; 24-hour deduplication.
• Custom vendors — config.extra_vendors = { "my-api" => "api.myservice.com" } adds private APIs to the dashboard.
• Multi-environment — Rails.env is captured automatically; production and development tracked separately.
• Stripe billing — Trial, Solo, and Team plans; Stripe-hosted checkout and billing portal.
• API key management — generate, list, and revoke keys from the dashboard; one-time plaintext reveal at onboarding.

SDK

• gem 'apidepth', '~> 0.1' — first RubyGems publish.