Changelog

New: capture a page without an account

• getsnap.dev/tools/screenshot.html — paste a URL, get a PNG. No account, no card, no extension.
• Six successful captures an hour per visitor. Only successful ones count: a typo, an unreachable site or a blocked address costs you nothing, and repeat URLs are served from cache.
• Fixed at 1280×720 PNG with no options. Everything else — full page, PDF, video, custom viewports, authenticated pages, scheduling, batch — stays in the API, where the free tier is 100 captures a month.

Security: capture URLs are checked against their resolved address

• Until now only the URL's syntax was validated, which accepts anything the URL parser understands. The renderer could be pointed at file://, at localhost, or at a private or link-local address including the cloud metadata endpoint.
• Every capture URL is now resolved and judged by the addresses it points to, not by how the hostname reads — a public name resolving to 127.0.0.1 is refused. Redirects are re-checked, because a redirect is a second navigation.
• Behaviour change: capturing an intranet address directly now returns 400 url_not_allowed. It was never a supported path — it worked only because the server could route there. To reach a private network, use a proxy.

Fixed

• Rejected requests no longer reach route handlers. Authentication replied correctly and the handler ran anyway, which showed up as a status code that disagreed with its own response body on POST /v1/render-link. Traced to a framework behaviour, reported upstream, and written up.
• Horizontal scrolling on phones. Several pages were wider than the viewport, which pushed the centred content off to one side — worst on the homepage, where a decorative element extended 115px past the left edge.
• Links inside paragraphs and lists are underlined again, so they are not distinguished from surrounding text by colour alone.
• The homepage no longer downloads all six demo captures on load. It fetches the visible one and the next, which cut about 1.2MB from first load.

We no longer store your API key

• Only a SHA-256 hash of your key is kept. It is enough to recognise the key you send, and cannot be turned back into the key.
• Your key is shown once — in the POST /v1/signup response and in your welcome email. Nobody, including us, can retrieve it afterwards. Lost it? Email support@getsnap.dev and we will revoke it and issue a new one.
• Why it matters beyond the obvious: the database, and every nightly backup of it, previously contained live credentials. Now they contain hashes.
• SHA-256 rather than a password hash is deliberate. Slow hashes exist to defend low-entropy human-chosen secrets against offline guessing; an API key is 32 random characters, so there is nothing to guess, and a slow hash would be paid on every authenticated request.
• Nothing to do. Existing keys keep working exactly as before.

Breaking: render links signed before this release stop working

• Regenerate them with POST /v1/render-link. The URL format is unchanged — still key=pk_… plus a signature — so nothing in your code needs to change.
• v1.15.0 moved the secret out of the URL but still keyed the signature with a value stored on your account row. Anyone able to read the database, or a backup of it, could therefore mint valid links for any account and burn its quota.
• The signature is now keyed by a secret held only in the server environment, combined with your account. Reading the database no longer yields anything that can sign.
• There is deliberately no grace period accepting old signatures: honouring them would preserve exactly the weakness this removes.

Fixed

• Rejected requests no longer reach route handlers. Replying from an authentication hook stopped halting the request chain once the app registered a second onSend hook — a Fastify behaviour we reproduced in isolation and reported upstream as fastify/fastify#7090. Visible symptom was a status code that disagreed with its response body on POST /v1/render-link.
• Four unhandled promise rejections that could terminate the API process: a welcome email sent without a catch, an async shutdown handler, an async screencast frame listener, and a selector failure that discarded its original error.
• The API no longer serves the marketing site, so api.getsnap.dev can no longer answer with a duplicate of a page that belongs on getsnap.dev.

Breaking: render links use a public handle

• POST /v1/render-link now returns a URL containing key=pk_…, a non-secret handle for your account, instead of api_key_id=sk_live_….
• Links generated before this release stop working. Regenerate them. The change is deliberate rather than incidental: /v1/render-link exists so you can put a capture in an <img> tag without exposing a key, and it was doing the opposite — publishing the secret into page source, Referer headers, CDN logs and ours. Because the signature is HMAC'd with that same secret, anyone holding one link could mint any other.
• The signature is still keyed by your secret, so only you can create a valid link. The secret simply never leaves the server now.
• Every account gained a public_id, backfilled automatically. Nothing to do beyond regenerating links.

Fixed

• Scheduled captures that omitted format failed on every run, with page.screenshot: options.quality is unsupported for the png screenshots. A schedule created as {"url": "…"} — the obvious thing to create — never once succeeded. The worker now applies the same defaults a direct /v1/screenshot call gets.
• Team members could be charged for nothing. /v1/upgrade had no notion of teams, so a member of a free team could buy Starter, have it charged, and see no change — because effective limits come from the team row. It now returns 409 team_billing_required and points owners at /v1/teams/:id/checkout.
• Cache keys ignored options that change the capture. lazy_load, wait_for_selector and wait_for_network_idle were absent, so a request asking for lazy_load could be served an image taken without it. Also added fail_if_selector_missing and proxy credentials. Existing cache entries are unaffected.
• A missing x-api-key returned 400 describing the request body, and on routes with required body fields it reported those instead, hiding the auth failure entirely. Now 401 unauthorized with a message naming the header.
• Validation errors name the part that failed — Invalid request headers / querystring / params — rather than always claiming body.
• Favicon corners rendered as white pixels on dark browser tabs; the raster copies had been flattened without an alpha channel.

Changed

• POST /v1/signup returns verification_required, and its message no longer claims the key is ready to use. With email verification enabled the key returns 403 email_not_verified until the activation link is clicked, which the response and the docs now both say.
• Stripe Checkout metadata carries your public handle instead of your live API key.
• usage_log rows are deleted automatically after 90 days, which the privacy policy had always claimed and nothing enforced.
• Logs redact any sk_live_ / sk_test_ token before writing.

Added

• captures_total is now actually emitted on /v1/metrics, labelled by kind, status and cache. It had been declared and never incremented, so the counter read as a flat zero.
• /v1/upgrade, /v1/billing and /v1/signup document their full response set in the OpenAPI spec, including the new 409.
• Per-language quick starts for Python, Node.js, PHP, Ruby, Go, Java and C#, and hosted-vs-self-hosted comparisons for Puppeteer, Playwright, Selenium and html2canvas.

Runtime alerts (Discord)

• New src/services/alerts.ts service fires color-coded Discord embeds for three classes of prod incident:
  • Stripe webhook signature failure - 1h Redis dedup. Fires when someone posts to /v1/webhook/stripe with a missing / rotated / spoofed signature header. Muted so a bot sniffing the endpoint can't spam the channel.
  • API 5xx response - 10min Redis dedup (env-tunable via ALERT_5XX_DEDUP_SECONDS). Fires on the first 5xx per rolling window with request_id + method + path + error so on-call can grep the droplet logs.
  • Cron script fatal error - no dedup. sendDrips / backupDb / restoreDb all alert on top-level exceptions. The shell wrappers (hourly-drip.sh, daily-backup.sh) also POST to Discord on non-zero exit to catch docker exec-level failures the Node script can't self-report.
• Wired via Fastify onSend hook (catches all 5xx including explicit reply.code(500) from route handlers, not just uncaught exceptions).
• All alerts are fire-and-forget from the call site; a broken webhook or Redis outage cannot mask the underlying error being reported. Fail-open on Redis errors (prefer noisy to silent). No retry on Discord POST (5s timeout, drop over back-up).
• 10 new tests covering dedup + payload shape + failure resilience. See ops/ALERTS.md for the triage runbook.

Deploy notifications (GitHub Actions)

• New "Notify Discord" step in .github/workflows/deploy.yml fires for every deploy (success / failure / cancelled) with color-coded embeds. Failure variant says "auto-rolled back"; title links to the workflow run URL for one-click log access.
• CI failures ping Discord too now (extension of the deploy pattern to .github/workflows/ci.yml).

Full-site WCAG 2.1 AA sweep

• Audited all 26 user-facing pages via the /v1/audit endpoint (dogfooding). Found 18+ violations across 8 pages; fixed all of them.
• Landmark fixes: <main> added to 11 pages (changelog + 10 blog posts + docs/playground). Fixes landmark-one-main + region violations that flagged 20-30 nodes each.
• Heading order: footer h4 → h3 on landing; release-body h4 → h3 on changelog; "Pricing Comparison" h3 → h2 on all 4 compare pages.
• 50 unassociated <label> tags on /docs/playground.html gained for= attributes.
• Color-contrast: every color:var(--primary) used as text bumped to var(--primary-text). method.delete gained --red-light.
• Link-in-text-block: global underline rules for .footer-links a, .article p/li/td a, changelog paragraph links, .docs-content h3 a.
• Scrollable-region-focusable: new landing/a11y.js (250B gz) referenced from all 26 pages adds tabindex="0" to every <pre> so keyboard users can arrow-scroll code blocks.
• Result: 0 violations across all 26 pages.

Documentation completeness

• README.md endpoints table went from 19 rows to 30, grouped into 8 categories.
• /docs/api-reference.html gained a "Complete endpoint index" section listing every operation with links to detailed sections or the interactive Swagger UI.
• New blog post: Team-owned Stripe subscriptions in 3 endpoints - case study of shipping v1.13 team billing without a per-seat model.
• New runbook: ops/ALERTS.md with per-event triage playbook + Redis DEL commands for forcing re-fire during triage.

SDKs (getsnap@1.14.0)

• Python: added __version__ module attribute (was missing since v1.0). Now getsnap.__version__ returns "1.14.0".
• No API surface changes - identical to 1.13.0 for both npm + PyPI. Bump is for cache invalidation + documentation timestamp.

Config

• DISCORD_ALERT_WEBHOOK_URL + ALERT_5XX_DEDUP_SECONDS added to .env.example and documented in src/config.ts zod schema.

Regression check

• Test suite: 254 → 264 tests, all passing (10 new alert tests).
• Live smoke test: all 12 API paths return expected shapes; auth wall correct; team billing endpoints route through metadata.team_id correctly (verified via direct Stripe API subscription create with metadata + observing team row update).
• All 25 pages audited: 0 WCAG violations.

Team billing

• Teams now have their own Stripe subscription, decoupled from the creator's personal billing. Previously, at team-creation time the team copied the creator's stripe_customer_id / stripe_subscription_id and any downstream Stripe events only ever touched the creator's api_keys row, never the team — the team's plan silently drifted whenever the creator up/downgraded.
• The owner now runs POST /v1/teams/:id/checkout which creates a fresh Stripe customer for the team, launches Checkout with metadata.team_id stamped on the resulting subscription, and returns the checkout URL. New team subs get a 7-day trial with missing_payment_method: cancel.
• Downstream webhook events (customer.subscription.created / updated / deleted / trial_will_end) check subscription.metadata.team_id and route to the team row via new updateTeamPlan / downgradeTeamToFree service functions. Redis cache is punched for every attached api_key so members see the new limits instantly, not on TTL.

API endpoints

• POST /v1/teams/:id/checkout — owner-only. Body {plan: starter | pro | business}. Returns {url, type: "checkout" | "portal"}; if the team already has an active sub, punts to billing portal.
• POST /v1/teams/:id/portal — owner-only. Stripe billing portal for the team's subscription. Returns 400 if no subscription exists yet.
• GET /v1/teams/:id/billing — any team member. Returns plan, limits, subscription_status (active / trialing / past_due / canceled / unpaid / incomplete / null), current_period_end, cancel_at_period_end, has_team_subscription.
• Total endpoints in Swagger: 27 → 30. Team paths: 7 → 10 (11 operations).
• DELETE /v1/teams/:id now also cancels the team's Stripe subscription before dropping the DB row (fire-and-forget with error logging) so owners aren't charged after dissolving.

Schema migrations

• Three new columns on teams via idempotent ALTER TABLE: subscription_status TEXT, current_period_end TEXT, cancel_at_period_end INTEGER DEFAULT 0.
• POST /v1/teams no longer copies the creator's stripe_customer_id or stripe_subscription_id. New teams start with those fields empty; existing teams keep whatever inherited values they already have (owner can re-run /checkout to migrate).

SDKs

• getsnap@1.13.0 on npm and PyPI. Node: snap.teams.checkout(teamId, {plan}), snap.teams.portal(teamId), snap.teams.billing(teamId). Python: snap.create_team_checkout(team_id, plan), snap.get_team_billing_portal(team_id), snap.get_team_billing(team_id).
• CLI: getsnap teams checkout <team_id> <plan>, getsnap teams portal <team_id>, getsnap teams billing <team_id>.

Regression check

• Test suite: 246 → 254 tests, all passing. 8 new tests cover metadata-based routing, empty-metadata fallthrough, subscription_details.metadata fallback path, team dedup, owner-email lookup for trial_will_end.
• Live end-to-end smoke: created a test team → ran /checkout (fresh cus_... created with team metadata) → directly created a subscription via Stripe API with metadata.team_id → verified webhook fired and updated teams.plan, monthly_limit, rate_limit_per_min, subscription_status, stripe_subscription_id. Cancelled the sub → team status transitioned correctly.

API

• POST /v1/teams/:id/transfer-ownership — owner promotes a member to owner and gets demoted to admin. Every team has exactly one owner at all times; this is the only way to change ownership without dissolving. Body: { new_owner_user_id }.
• DELETE /v1/teams/:id — owner permanently dissolves the team, detaches every member's api_key back to their personal plan, revokes outstanding invites, drops the team's monthly_usage row. Returns { detached_key_count, invites_revoked }. One-way action, no undo.
• Total endpoints in Swagger: 25 → 27. Team routes now cover the full lifecycle owner → admin transitions without support tickets.

Usage rollover on team creation

• When a solo user creates a team, their current-month personal monthly_usage row is now copied into the team's counter as part of the create transaction. Previously usage reset to zero when a solo Pro user made a team on the 20th of the month, effectively giving them a fresh 25,000 quota. Fixed.

SDKs

• getsnap@1.12.0 on npm and PyPI. Node: snap.teams.transferOwnership(teamId, { new_owner_user_id }) + snap.teams.dissolve(teamId). Python: snap.transfer_team_ownership(team_id, new_owner_user_id) + snap.dissolve_team(team_id).
• CLI: getsnap teams transfer <team_id> <new_owner_user_id> + getsnap teams dissolve <team_id> --yes. Dissolve requires an explicit --yes flag because it's irreversible.

Docs

• /docs/teams.html and /docs/referrals.html are now full reference documentation (previously just interactive dashboards). Teams doc gained a permissions matrix, all 8 endpoint references with curl / Node / Python / CLI samples, a data model, and a 10-question FAQ. Referrals doc gained reward-mechanics tables, cookie attribution deep-dive, anti-abuse rules, and an expanded 8-question FAQ.

SDKs

• getsnap@1.11.0 on npm and PyPI — adds the snap.teams.* namespace (Node) / snap.create_team() et al. (Python) wrapping all six team endpoints with typed request / response shapes.
• CLI adds a getsnap teams family: list, create, members, invite, accept, leave. Terminal-friendly output; all subcommands accept --json for machine-readable output.

Accessibility

• Ran our own POST /v1/audit against getsnap.dev and found 60 WCAG AA color-contrast violations, all stemming from three color patterns (primary-on-dark text, white-on-primary buttons, red 'no' cells in the comparison table).
• Fixed by splitting --primary into two roles: --primary (#4f46e5) for backgrounds and borders where white text sits on top; --primary-text (#a5b4fc) for text on dark backgrounds. Plus --red-light (#fca5a5) for the comparison table crosses.
• Re-audit: 60 → 0 violations. Full WCAG AA sweep now returns zero total violations.

Bug fixes

• Team-member api_keys were returning 500 FOREIGN KEY constraint failed on every capture because monthly_usage.api_key_id had a FK to api_keys.id, but the teams refactor started routing usage under the team ID (which doesn't exist in api_keys). Fix: inline migration drops the FK constraint (table-swap since SQLite has no ALTER DROP CONSTRAINT).

Six new endpoints

• POST /v1/teams creates a team, hoists your current plan/quota/rate onto it, and attaches your api_key
• GET /v1/teams lists teams you belong to with your role in each
• GET /v1/teams/:id/members lists the roster with emails and roles
• POST /v1/teams/:id/invites invites a member by email (owner/admin only). Fires a branded invite email with a single-use link that expires in 7 days.
• POST /v1/teams/accept consumes an invite token, adds the caller to the team, and attaches their api_key to the shared plan
• POST /v1/teams/:id/leave lets non-owners leave the team; api_keys detach and fall back to personal limits

How the shared pool works

• All requests from any team member's api_key deduct from ONE shared monthly quota and one 60-second rate-limit window.
• Individual api_key logging still tracks each user's usage inside usage_log, so you can see who made what.
• Team members keep their own api_keys — no key sharing, no unrevealed secrets.
• Solo users are unaffected: api_keys.team_id defaults to NULL and everything works as before.

Roles

• owner — created the team. Cannot leave via /leave (transfer or dissolve instead). Every team has exactly one.
• admin — can invite / remove members, cannot dissolve.
• member — can view team info and use the shared quota.

Anti-abuse

• Invites verify caller's email against the invited email before accepting
• Cannot re-invite an existing member (returns 409 already_member)
• Owners cannot leave via /leave endpoint
• Role escalation via double-invite is blocked (existing membership takes precedence)

New endpoint

• POST /v1/audit runs axe-core against any URL and returns categorised WCAG 2.0 / 2.1 / 2.2 violations grouped by impact (critical, serious, moderate, minor).
• Every violation includes the rule ID, plain-English description, WCAG success criterion tags, and the failing HTML elements with CSS selectors so you can jump straight to the fix in your codebase.
• Optional tags filter (wcag2aa, wcag21aa, wcag22aa, section508, ACT, etc.) narrows the audit to a specific conformance level.
• Optional rules filter (e.g. color-contrast, label, region) runs only specific axe rules for targeted CI gates.
• Optional include_screenshot flag captures a viewport screenshot alongside the report (+1 credit) so you have visual context in review.

CLI

• New getsnap audit <url> command with a color-coded, terminal-friendly summary. Shows impact tier counts, top violations with rule ID + help + deque.university docs link, and a status marker (!!, !, *, OK) at the start of each line for CI log skimming.

SDKs

• getsnap@1.10.0 on npm and PyPI — adds typed audit() methods with full AuditOptions / AuditResponse / AxeViolation shapes.

Pricing

• 1 credit per audit + 1 more if you request a screenshot. No cache (DOM state is per-invocation) — but audits are 3-5x cheaper than a full capture on average since we skip the heavy image encoding.

New endpoint

• POST /v1/diff captures two URLs at identical viewport dimensions, compares them pixel-by-pixel with pixelmatch, and returns a similarity score (0-100) plus three CDN URLs: before, after, and a red-overlay diff image.
• Both captures run in parallel to halve latency. Cache HITs on either side reduce the credit charge - a diff of two pages already in cache costs 0 credits.
• Per-side overrides for wait_for_selector, extra_delay_ms, and css handle sites where each side settles at different rates.
• pixelmatch tunables exposed: threshold (0-1 YIQ sensitivity), anti_alias, include_aa. Sensible defaults for typical marketing / product pages.

CLI

• Every SDK install now ships a getsnap executable. No code required for one-off captures:

export GETSNAP_KEY=sk_live_...
npx getsnap screenshot https://news.ycombinator.com --out hn.png
npx getsnap pdf        https://news.ycombinator.com --out hn.pdf --full-page
npx getsnap og         https://blog.example.com/post --out og.png
npx getsnap video      https://getsnap.dev --out demo.mp4 --duration 5
npx getsnap diff       https://staging.mysite.com https://mysite.com
npx getsnap usage
npx getsnap referrals

Every capture flag from the SDK is available on the CLI: --full-page, --viewport 1440x900, --wait-for #hero, --dark-mode, --device mobile, etc. Full reference: npx getsnap --help.

SDKs

• getsnap@1.9.0 on npm and PyPI — adds typed diff() methods with full DiffOptions / DiffResponse shapes. Node package now ships the CLI too via a bin entry.

New endpoint

• GET /v1/referrals returns your unique referral code, ready-to-share URL, reward tier, and aggregated stats (total, paid, pending, reverted, lifetime earnings in cents).

How it works

• Every account gets an auto-generated 8-character referral code (uppercase alphanumeric, ambiguous chars like 0/O/1/I/L excluded).
• Share https://getsnap.dev/?ref=YOURCODE. Landing JS captures the code into a 30-day first-party cookie on the referee's visit.
• When the referee signs up, the code follows them into the request. On their first paid checkout (Starter, Pro, or Business), we attach a Stripe coupon giving them a discount on their first month.
• When the referee's first invoice is actually paid (i.e. they finished the trial + kept the plan), you get a $5.00 credit added to your Stripe customer balance.

Anti-abuse

• Self-referral (same email on both sides) is silently ignored.
• Each referee can only be attributed to one referrer, ever, enforced by a UNIQUE constraint at the DB level.
• Reward fires only on the FIRST successful invoice - trial cancellations or later refunds don't burn credits.
• Reward is idempotent under Stripe webhook retries and repeat-billing invoices for the same subscription.

New endpoint

• Added POST /v1/og-image: a convenience wrapper over /v1/screenshot with defaults tuned for social share images — 1200×630 PNG, hi-DPI (2x), ad and popup blocking on, waits for network idle.
• Same billing as /v1/screenshot: 1 credit per fresh capture, 0 for cache hits.
• Supports response_type: "url" (JSON with CDN URL, default) or "binary" (raw image bytes with Cache-Control: public, max-age=604800, immutable).

Billing reliability

• Stripe webhook is now idempotent. Retried event deliveries (up to 3 days) no longer risk double-charging or double-applying subscription changes. Backed by a stripe_events UNIQUE-constrained log with transactional rollback.

Onboarding

• Day-3 activation email lands automatically for new signups who haven't made an API call yet. Real paste-and-run code (curl, Node.js SDK, Python SDK) instead of a marketing pitch.
• Trial-ending email fires 72 hours before your Stripe trial expires. Includes a personalised billing portal link so you can update or cancel in one click. Wired via Stripe's customer.subscription.trial_will_end webhook.

SDKs

• getsnap@1.6.0 on npm and PyPI — adds ogImage() / og_image() convenience methods with typed OgImageOptions / OgImageResponse shapes.

Ops

• Automated production deploys on git tag push. .github/workflows/deploy.yml checks out the tag, rebuilds via docker compose, polls /v1/health for up to 90 seconds, and auto-rolls back on failure.
• Restore script now supports RESTORE_TARGET_DB_PATH for safe drills against a scratch path. Documented RTO/RPO in ops/RESTORE-DRILL.md.
• Load test script + baseline results at ops/load-test.js and ops/LOAD-TEST-RESULTS.md. Verified production sustains 51 req/s at p95 <30ms with 12% memory utilisation.

API

• New proxy_url, proxy_username, proxy_password, proxy_country fields on /v1/screenshot and /v1/video
• Supports http, https, socks5, socks5h schemes. Credentials stripped before handing to Playwright and SHA-256–truncated in cache keys.
• Managed geo pool via PROXY_URL_TEMPLATE env var. When configured, clients pass proxy_country: "US" and we substitute.
• Optional PROXY_ALLOWED_COUNTRIES allow-list restricts which ISO codes are accepted.

Docs

• Swagger UI at api.getsnap.dev/docs now shows the full request body (up to 74 parameters), response shape, and every error code for each endpoint. Previously it rendered barebones stubs.
• The interactive playground at /docs/playground exposes every Phase 6–11 field in collapsible sections: proxy, PDF metadata, extract_metadata, clip, slicing, blocking, emulation.

SDKs

• getsnap@1.5.0 on npm and PyPI — adds proxy_* fields to ScreenshotOptions and VideoOptions, plus HMAC verification example.

New endpoint

• POST /v1/video records a smoothly-scrolling capture as MP4 (libx264 + faststart, quality-driven CRF) or GIF (two-pass palettegen → paletteuse).
• 10–30 fps, 2–30 second duration, viewport up to 1920×1080.
• Uses Chrome DevTools' Page.screencast protocol for high-fps frame sampling (not a screenshot loop) and ffmpeg for encoding.
• Sync / async webhook / raw binary response modes. Async webhooks include a video.completed event.
• Costs 1 credit per second of video (rounded up, min 1).

Infrastructure

• Dockerfile now installs ffmpeg alongside curl.
• New SCHEDULE_WORKER_CONCURRENCY env var so cron-driven captures don't starve the sync pool.

SDKs

• getsnap@1.4.0 Node: client.video(), client.videoBinary(), full VideoOptions types.
• getsnap@1.4.0 Python: gs.video(**kwargs), gs.video_binary(**kwargs). First release published to PyPI.

New endpoints

• POST /v1/schedules, GET /v1/schedules, GET /v1/schedules/:id, PATCH /v1/schedules/:id, DELETE /v1/schedules/:id
• POST /v1/schedules/:id/run — trigger a schedule immediately for testing
• GET /v1/schedules/:id/runs — execution history (last 100 runs kept)

Design

• BullMQ repeatable jobs, Redis-backed, survive restarts via startup DB hydration
• IANA timezone support (validated with cron-parser)
• Optional webhook_secret triggers HMAC-SHA256 delivery signing: X-Webhook-Signature: sha256=<hex> and X-Webhook-Timestamp so receivers can verify authenticity and defeat replay attacks

SDKs

• Node: client.schedules.{create, list, get, update, remove, runNow, runs}
• Python: create_schedule / list_schedules / get_schedule / update_schedule / delete_schedule / run_schedule_now / list_schedule_runs

Category A — new capture behaviour

• clip: { x, y, width, height } — capture a specific rectangle of the viewport, overriding full_page and selector
• full_page_max_height — hard cap on infinite-scroll pages
• full_page_slices, full_page_slice_height, full_page_slice_overlap_height — split tall captures into overlapping strips for AI vision workflows
• reduced_motion and media_type: "screen" | "print" emulation
• click_accept — auto-click cookie-consent Accept buttons (OneTrust, Cookiebot, FundingChoices, generic heuristics)
• press_escape — dismiss modals via Escape keypress
• fail_if_selector_missing — hard-fail on missing selectors instead of silently ignoring
• attachment_name — sets Content-Disposition on binary responses
• AVIF as a first-class output format

Category B — PDF completeness

• Paper sizes: A0–A6, Letter, Legal, Tabloid, Ledger
• Orientation, per-side margins (top/right/bottom/left), pdf_background, pdf_scale
• Templated pdf_header / pdf_footer (Chromium format: <span class="pageNumber">, totalPages, title, url, date)
• Page range filter (pdf_page_range: "1-5,8")
• Document metadata: pdf_title, pdf_subject, pdf_author, pdf_keywords, pdf_creator

Category C — extraction

• extract_metadata returns {title, favicon, open_graph, fonts, http_status, http_status_text, http_headers} in one field

Category D — request blocking

• block_images, block_fonts, block_scripts, block_frames, block_urls (wildcard patterns), bypass_csp

Dependencies

• sharp for AVIF conversion + full-page slicing
• pdf-lib for PDF metadata post-processing

• New brand across the entire stack: landing, docs, compare pages, SDKs, email templates, Swagger title, webhook User-Agent
• Class renamed SnapAPI → GetSnap in the Node SDK; backwards-compat aliases preserved (SnapAPI, SnapAPIError)
• New comparison landing table with 6 columns and honest research-based rows for ScreenshotAPI, Urlbox, ScreenshotOne, ApiFlash
• Four dedicated /compare/*.html deep-dive pages
• Node SDK getsnap published to npm (renamed from a legacy internal name)

Core API

• POST /v1/screenshot — screenshot or PDF of any URL or HTML string
• POST /v1/batch — up to 100 URLs in one request
• POST /v1/render-link + GET /v1/render — signed GET URLs for <img> embedding
• GET /v1/usage, GET /v1/health

Capture features

• AI popup removal (90+ heuristic selectors), ad blocking
• Full-page, mobile/tablet emulation, dark mode, Retina/HiDPI
• Element capture via selector, transparent PNG
• Wait-for-selector, lazy-load auto-scroll, click/scroll before capture
• Custom CSS/JS injection, cookies, headers, locale, timezone, user agent
• Text and HTML extraction
• Async webhooks, S3 upload to user-owned buckets

Platform

• Fastify + Playwright + Redis + SQLite stack on a single droplet
• DigitalOcean Spaces CDN with hash-based caching
• Sliding-window rate limiting, monthly quota enforcement
• Stripe metered billing with 4 plans (Free / Starter / Pro / Business)