What we shipped.

Onboarding that finishes the job

A live walkthrough with fresh tenants drove a full onboarding overhaul: the concierge now completes your profile instead of collecting answers and dropping them — and signing up without a website, or without an AI key, is a first-class path.

• IMPROVEDThe concierge derives your product category — the field the whole adaptive-content system keys on. Deterministic keyword match first, AI classify as fallback, and it never overwrites a category you picked by hand.
• IMPROVEDLocal means local — an optional town/city question feeds weather context and genuinely local landmark suggestions instead of national clichés. No city? Suggestions stay country-level and say so.
• NEWNo-website signups — Instagram-only and pre-launch brands can now sign up without a URL. The brand-voice scan skips honestly rather than inventing a voice from thin air.
• IMPROVEDEvery new tenant gets its 14-day trial seeded at creation (and backfilled at boot for tenants that pre-date it), so the day billing switches on nobody gets locked out of their own account.
• IMPROVEDSetup step outcomes persist: a blank brand voice after setup becomes a standing dashboard next-action instead of a stream message that scrolls away forever.

Your keys, your money — and a studio that fits any product

The tenant-readiness audit wave: BYOK enforced on every AI surface, category-neutral content for non-drinks brands, and a mobile/flow pass on the editors.

• FIX18 AI surfaces now bill the right account — several modules (SEO blog, campaigns, reviews-to-UGC, support inbox, competitive intel and more) were gated on the platform's own key instead of yours. Every one now resolves your connected key, and image generation — the costliest class — is strictly tenant-keyed.
• IMPROVEDCategory-neutral content — dozens of prompts that quietly assumed a drinks brand now derive their scenes, subjects and angles from your actual product category.
• IMPROVEDMobile + flow pass: editor layouts that work on a phone, unsaved-edit warnings, and no dead-end screens.
• IMPROVEDSignup derives your currency and locale from country + browser language instead of defaulting everyone to GBP.
• NEWEmail Studio goes prompt-first: describe the email you want and the AI drafts the full section plan; every section has a one-click AI fill.

Email Studio — a real email builder, honestly wired

Nine phases shipped in one arc: a visual section builder that compiles validated email HTML, with AI planning, your real product catalogue, guarded imagery, and version history. The old form-based email path is retired.

• NEWVisual section builder — compose from typed sections (hero, products, story, footer…) with a three-panel editor, brand profile styling and 16 starter templates.
• NEWAI campaign planning — plans a campaign and writes per-section copy suggestions you accept or edit; nothing publishes itself.
• NEWYour real catalogue — product sections pull live Shopify data with drift detection, so a price change flags the email that still shows the old one.
• NEWExport for Shopify Messaging — compiles validated HTML for Shopify's editor via a guided paste flow. (Shopify has no draft-email API; we say so in the product rather than pretending.)
• NEWVersion history with one-click duplicate-as-draft, plus an expert override for hand-tuned HTML.
• IMPROVEDThe classic form-based email builder is retired — one email path, no conflicting token systems.

Events & Markets — listings researched, verified and published

A full events system for local-market brands: AI research with named sources, human review, Shopify publishing, a storefront section, and monitoring that re-checks official sources as event day approaches.

• NEWAI event research with receipts — every field carries its source and a confidence score, and anything conflicting or uncertain routes to human review instead of publishing.
• NEWEvent imagery with rights tracking — uploads and gated generation, plus auto-generated 3:4 posters when no usable image exists.
• NEWShopify publishing — events publish as metaobjects with an on-brand storefront theme section (live-proven end to end).
• NEWSource monitoring — published events get their official sources re-checked on a distance-based cadence (weekly far out, daily close in); material changes to dates or venue pull the listing back to review with old value, new value and source shown. Nothing is auto-overwritten.
• NEWAutomation modes from fully-manual to fully-automatic — and even "fully automatic" refuses to publish an event it can't identify confidently.

Judge creative against your numbers, in an interface that stays out of the way

The leaderboard now ranks your whole account and calls fatigue against your own baselines — plus the late-July UI wave: one floating dock, honest navigation, and the 390px mobile audit fixes.

• IMPROVEDLeaderboard ranks the account, not an arbitrary sample — and a grid view shows creative at a size you can actually judge.
• NEWOne fatigue view — creative fatigue judged against the tenant's own numbers, in one place, instead of scattered heuristics.
• IMPROVEDOne floating dock replaces four widgets fighting for a corner; the Engine's own button joins it.
• IMPROVED11 orphaned routes surfaced in navigation — and an orphaned route is now a build error, so pages can't silently vanish from the nav again.
• FIXThe four defects the 390px mobile audit found, fixed; a lost batch-amplify job is now recoverable.

The ladder sells autonomy — and the copy stops lying

Tier restructure to the current ladder (Solo £29 / Deputy £59 / Helm £99 / Agency £149) plus an honesty sweep: receipts for what the autopilot actually did, and the removal of every claim the product couldn't back.

• NEWDeputy and Helm — the £59 rung is renamed Deputy (Scarif drafts, you approve) and a new £99 Helm rung lets Scarif act inside guardrails you set, with API access included. Existing Studio subscribers resolve to Deputy automatically; the Founding 50 keep their terms.
• NEWReceipts — an in-app receipts panel and a weekly digest listing what ran autonomously, what it changed, and what it spent.
• FIXMarketing pages stopped selling credits, top-ups and tiers that no longer exist; every price on the site is pinned to the plan registry by a consistency test.
• FIXThe budget guardrail now checks the number the model actually asked for, not the one it assumed.

Security and reliability blitz

A pre-launch audit day: two remote-takeover doors closed, a data-eating profile save fixed, and backups proven by actually restoring one.

• SECURITY/api/setup/save required no authentication — a remote caller could overwrite a tenant's profile and mint an admin. Closed on both the tenant and apex hosts, and every tenant-scoped route now binds its session to its host tenant.
• SECURITYOperator credentials are no longer accepted in query strings, and query strings are redacted before they reach logs, disk or a model prompt.
• FIXSaving your profile deleted 40 of its 70 fields — including your AI key. Partial saves now merge instead of rebuild, and a blank answer means "I didn't answer", never "delete what you have".
• FIXBackups: the platform archive was silently missing from every backup; the health check said "configured" instead of "working". Both fixed and proven with a real restore.
• FIXThe onboarding wizard said "Connected" without saving anything; /create/ad-creatives was an infinite redirect loop in production; 15 dead links fixed — including every tile on a public sales page.
• IMPROVEDDeploys now gate on typecheck + the full test suite (CI had quietly stopped running five of its own suites).

Credentials encrypted at rest — including the ones already on disk

Four credential fields were missing from the encrypt list, and data written before the master key was set stayed plaintext. Both closed, plus an API-key bridge for machine clients.

• SECURITYanthropic/openai/firecrawl keys + the analytics token joined the encrypt-at-rest list, with a guard test that fails the build if a future integration adds a credential and forgets to list it.
• SECURITYA migration wraps credentials already written in cleartext — it refuses to run without the master key (so it can't report success while encrypting nothing), backs up each profile, and verifies the result.
• NEWConnect Claude — scoped API keys let machine clients (like Claude) call a bridge of read/generate tools against your account, gated by the same permission system as humans.
• IMPROVEDAd-platform write access (adOps) split from billing permissions — a team member can run ads without being able to touch the subscription.

Agencies get a console; clients get a portal

The agency arc: client-facing reporting with co-branded PDF export, client content uploads with consent tracking, and a self-serve agency signup — alongside the tier restructure that created the £149 Agency rung.

• NEWClient reporting portal — clients log in by magic link and see their own brand's KPIs (revenue, spend, blended ROAS, AOV) read-only, plus a co-branded PDF export carrying the agency's name and colours.
• NEWClient uploads — clients contribute their own photos/footage direct-to-storage with consent and provenance recorded; the agency views and downloads them from a cross-client library.
• NEWSelf-serve Agency signup: sign up, get provisioned, log in by magic link — no sales call required.
• SECURITYMagic-link generation now builds URLs from the trusted tenant domain — a spoofed Host header can no longer poison a client's sign-in link.
• NEWThe dashboard shows a "what Scarif knows about you" brand-memory card — the intelligence the engine uses, visible and correctable.

The Engine can see — and video learns to tell a story

Engine chat became a visual, glass-box copilot, and the video studio gained true story mode: continuity-chained scenes stitched into one narrated film.

• NEWEngine chat shows real creative — your ads render as visual cards in the conversation, and the model sees the actual ad images (not just their captions) when it critiques or ideates.
• NEWGlass-box applies from chat — the Engine proposes changes it never applies itself: live preview, verify, one-click undo. A suggestion becomes a prefilled Studio draft in one click.
• NEWStory mode — multi-scene videos with visual continuity chained scene to scene, stitched into one film with text overlays and a single consistent narrator (your ElevenLabs voice, your key).
• NEWMulti-product videos: companion products ride into scene generation as real visual references, not name-drops.
• FIXVeo failures now tell the truth: Google retired the old model ids (fixed), and "blocked" videos turned out to be real-people name checks — names are scrubbed and the real block reason is shown.

The front door: concierge onboarding + Shopify App Store readiness

One onboarding spine instead of two overlapping wizards, and the install flow hardened for merchants arriving from the Shopify App Store.

• NEWConcierge-first onboarding — one front door: answer a few questions and the concierge scans your site, drafts your brand voice and sets up the workspace. No more double-wizard.
• IMPROVEDLanding page rewritten to sell the product that exists today, not last year's.
• FIXShopify App Store install flow hardened: auto-created accounts verified, OAuth-callback failures surface on a public error page instead of a blank screen, and logged-in merchants on the app host route straight to their workspace.
• IMPROVEDMarketing Engine overhaul closed out: creative board with Core/Testing launch presets and Google-side completeness.

Pilot — the Google Ads engine joins the fleet

Scarif now flies both ad platforms. Google Ads gets the full treatment: daily account audit, one-click paused campaign pushes, negative-keyword mining, real-revenue conversion uploads, and auto-pilot coverage — everything the Meta engine does, adapted for Search + Shopping.

• NEWOne-click Google campaign push — a plain-English goal becomes a full paused Search campaign: tightly-themed ad groups, the search phrase pinned as headline 1, message-matched landing pages from your real product URLs, negative keywords attached, presence-only geo targeting, search partners off. Shopping campaigns push too when Merchant Center is linked.
• NEWDaily Google account audit — miscast conversion goals (page-views set as primary quietly poison Smart Bidding), search terms burning spend with zero conversions (LLM-classified: job seekers, DIYers, competitors), profitable campaigns losing impressions to budget, and Quality Score penalties. Each finding arrives as a recommendation with one-click Apply.
• NEWConversion truth — real Shopify purchases upload to Google via the modern Data Manager API (click-matched where a Google click id survived to the order; hashed-email enhanced conversions otherwise). Created secondary so it never double-counts your GA4 import.
• NEWRSA refresh loop — Google labels every headline; ads with 2+ LOW assets get a refresh recommendation whose Apply regenerates only the weak lines, keeps the winners, swaps the ad and pauses the old one.
• NEWCustomer Match with a hard policy gate — restricted categories (alcohol, gambling…) are refused outright, protecting your account. Non-restricted stores sync hashed customer emails in one click.
• IMPROVEDAuto-pilot now covers Google budget moves under the same guardrails as Meta: ±15% clamps, pacing checks, full audit trail.

The Meta engine goes end-to-end — and your tracking page stops lying

A single day shipped the entire Meta roadmap: in-app campaign controls, the creative leaderboard with the amplify/fatigue flywheel, Scarif Auto's evidence engine, research-grounded demographic playbooks, funnel-wide CAPI, competitor intel, and pacing enforcement. Plus an honest rebuild of tracking attribution.

• NEWCreative leaderboard (Performance → Ad leaderboard) — every ad ranked with real thumbnails, 14-day trends and Ads-Manager deep links. ⚡ a winner to seed five more like it; fatigued ads flag themselves for regeneration.
• NEWScarif Auto — the studio's auto mode now reads four evidence sources (model P&L, audience segments, winning hooks, approval history) plus research-grounded playbooks for your demographic and product category, then directs generation at your best-converting audience. Its first catch on a live account: men 65+ converting at 6.6× vs a 4.4× blend.
• NEWAudience intelligence — personas become real Meta targeting in one click, saved audiences import, and audience P&L shows which segments actually pay.
• IMPROVEDHonest attribution — the tracking page now separates delivery (did your purchases reach the ad platforms?) from attribution (how many did ads drive?). Those are different questions; conflating them caused false alarms on healthy stores.
• IMPROVEDFunnel-wide server events: AddToCart and InitiateCheckout join Purchase in the Conversions API feed, with per-event delivery receipts.
• FIXPurchase webhooks registered via OAuth installs were 404ing on a route mismatch — fixed; delivery receipts now chart correctly.

Sovereign becomes the front door

Pricing pivot: the self-hosted, one-time-fee Sovereign licence (£1,999 + optional £499/yr maintenance) is now the headline offer, with hosted Solo £29 and Studio £99 alongside. The tier ladder from early May (Founder/Brand/Studio/Atelier) is retired.

• NEWSovereign licences are issued automatically: Stripe checkout → signed Ed25519 licence → email, in under a minute.
• IMPROVEDHosted plans simplified to Solo £29 / Studio £99 with a 20% annual discount. BYOK on every tier — your AI keys, zero markup, forever.

Phase 0.5 + Phase 1 + Phase 2 — Hormozi-validated offer foundation

Tier rename, bonus catalog, risk reversal, Founders Charter, lead magnets, distribution prep — plus the new homepage and pricing page that lift them all.

• NEWTier rename: Founder £19 / Brand £49 / Studio £149 / Atelier £399 (was Solo/Pro/Agency). Sovereign self-host tiered: £1,999 / £4,999 / £8,999. Annual discount 15% (industry standard).
• NEWFounders Charter — £499 lifetime hosted, capped at 100 customers. Atomic seat assignment, live counter on /founders-charter and homepage, sold-out → Charter II waitlist.
• NEWBonus catalog — every paid tier ships with stacked bonuses (£497-£4,996 stated value depending on tier). Brand voice playbook, 50 templates, Discord, migration assistant, 1:1 onboarding, quarterly review, monthly office hours, client reports, priority SLA, affiliate programme.
• NEWRisk reversal — 14-day money-back on hosted, 24-hour install guarantee on self-host. 'Tom personally guarantees' founder-presence component on every guarantee surface. Refund-request flow with auto-Stripe-refund inside the 14-day window.
• NEWAdd-on packs — extra brand £15/mo, +500/+1000 gens £29/£49, priority queue £19/mo, custom domain £15/mo. AddOnUpsellModal triggered at limit-hit.
• NEWLead magnets — public Brand Voice Generator at /tools/brand-voice (no-auth, AI-generated voice doc + 5 sample posts), £14K Stack Audit at /tools/stack-audit, Replacement Calculator at /tools/calculator.
• NEWAffiliate programme at /affiliates — 30% recurring 12-month commission. Application + Tom-review flow.
• NEWScarif Studio DFY waitlist at /services/studio — £1,999/mo managed service for £200K-£500K/mo brands.
• NEWMarketing homepage rebuild — 'Look like a brand 10x your size' hero, 6-workflow grid, comparison table, honest stack, customer story, pricing preview with bonus stacks, Founders Charter counter, guarantee banner, expanded FAQ.
• NEWPricing page rebuild — monthly/annual toggle (15% off), all 5 hosted tiers + 3 Sovereign tiers + maintenance + migration, add-on grid, comparison row, calculator CTA, pricing FAQ.
• NEWFree tier watermark composer (sharp-based) — applies 'scarif.one' watermark to gens on Try Scarif tier. Auto-removed when tenant upgrades.

Marketing site v1 — public pages live

Phase 47 of the production-launch roadmap: scarifone.com gets its real face.

• NEWLanding page at /marketing — confrontational hero, frustration → differentiation arc, 9-card feature grid, tier teaser, security pillars, FAQ, final CTA
• NEWFull /pricing page with comparison table covering every feature × every tier, top-up packs, plain-English FAQ
• NEW/about, /faq, /contact, /sovereignty, /trust — five public-facing pages all using a shared dark-mode marketing chrome
• NEW/legal/privacy, /legal/terms, /legal/dpa, /legal/cookies — UK GDPR-compliant baselines with plain-English summaries
• FIXRouting conflict between two /marketing pages was crashing the dev server
• IMPROVEDBrand colour CSS variables now flow through every public page — tenants on hosted plans see their brand on the dashboard chrome too

Phase 46 — multi-tenant in self-host + dedicated brand editor

The trial-week sprint that fixed every bug Tom hit during his 30-day customer-as-founder dogfooding.

• NEWSCARIF_ACTIVE_TENANT env var lets self-host operators run multiple tenant brands side-by-side and switch with a top-bar dropdown
• NEW/profile — dedicated brand-profile editor for existing tenants, separated from the first-run /setup wizard
• NEWAsset Library → Ad Creator workflow with contextual CTA, URL prefill, and one-click auto-fill from the asset
• FIXShopify connection state no longer shows false-negative on /create/ad-creatives when /integrations sees it as connected
• FIXConcierge SSE stream now propagates AI errors as proper exceptions instead of silently dropping them
• IMPROVEDAppShell — persistent sidebar + top bar across every authenticated page (was per-page chrome before)
• IMPROVEDDashboard redesign: full-width dark gradient hero, sectioned tile grid, brand-coloured tile borders

Trial-readiness sprint

Bug fixes + onboarding polish from the first round of customer trials.

• NEW9-integration Integration Hub with live-tested OAuth flows (Shopify / Meta / Klaviyo / Mailchimp / Judge.me / Google Ads / TikTok / Pinterest / LinkedIn)
• NEWSetup Concierge: 90-second autonomous AI agent that scans your store + drafts a 7-day content plan during onboarding
• NEW4 brand-voice setup paths (paste / upload / scan-website / sample-and-edit) — pick whichever fits your existing brand docs
• NEWBug-report widget on every page — every customer’s bug report gets fixed for everyone the same day
• IMPROVEDSetup wizard step 3 backspace bug fixed + every label rewritten for clarity
• IMPROVEDLocal dev runs out of the box — data path defaults to ./data when SCARIF_DATA_PATH is unset
• SECURITYPurged all personal references + tenant fixtures from the repository ahead of public release

Phase 41-43: spend caps, approval gates, model router

The three product pillars that make AI marketing safe to leave running unattended.

• NEWPer-feature spend caps with auto-pause — set per-day £5 and the system mathematically cannot spend more, even if a background agent goes wrong
• NEWApproval gates before every AI generation — see the cost estimate, review the prompt, approve or reject before tokens spend
• NEWFlash-first model router: cheaper models for routine work, Pro 2.5 only for the gens that need it. Per-tenant quality toggle for visual gens
• IMPROVEDQuality toggle (Standard / High) on ad-trends so you choose between cheaper-faster and slower-better per generation

Marketing engine: lifecycle, cart, reviews, SEO

The four highest-leverage marketing automations every ecommerce brand needs.

• NEWCustomer lifecycle segmentation (VIP / repeat / lapsed / at-risk) with AI-drafted email flows per segment
• NEWCart abandonment recovery: live abandoned-cart feed → auto-drafted 3-email sequence using the actual products in the cart
• NEWReviews → UGC pipeline: 5★ photo reviews auto-import → AI rewrites quote into 3 ad-ready variants → drafts permission email to customer
• NEWSEO + blog engine: AI long-tail post writer with Shopify push, schema.org generation, internal-link auto-suggestions
• NEWMarketing calendar — 200+ locale-aware seasonal events with AI campaign suggestions ahead of each one (BFCM, Mother’s Day, Diwali...)

Operations modules: subscriptions, wholesale, customer support

• NEWSubscriptions / recurring-revenue module — track and grow your recurring revenue, with AI-drafted retention sequences
• NEWWholesale + multi-currency: Shopify B2B catalog support with locale-tax-aware pricing (GBP / USD / EUR / AUD / NZD)
• NEWAI customer support inbox: drafts replies using your brand voice + customer’s order history. You review + send.
• NEWCompetitive intelligence: track competitor Shopify stores, weekly scans for new products + price changes + AI insights