arabface.ai
Partnership brief & 60-day plan — the MENA AI agents factory.
§0 TL;DR — read this first
What: a factory that ships one focused Arabic AI agent every 1–2 weeks under arabface.ai/<agent>. Bilingual Ar/En, MENA-positioned, globally accessible. Not a chatbot, not an LLM — a portfolio of narrow, purpose-built tools.
60-day test: ship two agents — photo restoration + Arabic podcast→shorts — and measure whether one produces a clear traction signal (viral + first D2C paying customers, OR first B2B paying studios).
Why now: generic Arabic chat is already won by sovereign free tools (HUMAIN / Fanar / Jais). But narrow, dialect-aware, workflow-specific agents for MENA consumers and SMBs are still mostly vacant. ~6–12 month window before someone else claims the "MENA AI agents factory" slot.
Decisions this week: name commitment, US LLC, capital ceiling, equity split, time commitment. See §8.
§1 The Bet
Thesis: the Arab/MENA AI tool market has two crowded extremes and an empty middle.
- Top — Sovereign LLMs, all free, state-backed: HUMAIN Chat (PIF/SDAIA, #1 in Saudi App Store), Fanar (Qatar/QCRI), Jais (UAE/G42). They own generic chat.
- Bottom — Enterprise platforms with long sales cycles: Lucidya ($30M Series B), Intella ($12.5M A), Arabic.AI/Tarjama ($15M A), WideBot, Arabot. They own enterprise contracts.
- Middle — Consumer/SMB narrow agents. Araby.ai is the lone real player (~$1M ARR, 11–20 staff). Wittify.ai is pre-seed ($1.5M) but enterprise-leaning. This is where we fit.
The wedge
Ship narrow agents that each solve one visible pain excellently in Arabic. Accumulate under one brand. Let the portfolio discover product–market fit instead of betting everything on one hypothesis. Pieter Levels / Marc Lou model + regional identity brand. That combo has no incumbent.
Brand thesis
"arabface" is figurative, not literal — the face of Arab AI on the web. That said, agent #1 happens to work with the literal meaning too, which is useful marketing insurance.
§2 Market & competitive evidence
What's already shipping
| Name | Category | Pricing | Signal |
|---|---|---|---|
| ChatGPT | Generalist | $0 / $20 | 91% MENA share (Jul 2025) |
| HUMAIN Chat | PIF/SDAIA chat | Free | #1 Saudi App Store |
| Fanar | QCRI multimodal | Free API | Free forever (state-subsidized) |
| Jais Chat | G42 / Inception | Free web | Researcher-oriented |
| Araby.ai | Arabic content suite (broad) | $8.99–$39.99 | Riyadh 2022. 11–20 staff, 1 engineer. Self-funded. ~$1M ARR, $3.3M valuation. Beatable — see §2.5 |
| Wittify.ai | Arabic agents | Pre-seed beta | $1.5M raised Jul 2025 — watch closely |
| Lahajati | Arabic TTS (192 dialects) | Freemium | 15k+ creators |
| Munsit | Arabic STT | Enterprise | <10% WER on dialects |
| Lucidya | Enterprise CX | $$$ | $30M Series B 2025 |
| Intella | Arabic speech | $$$ | $12.5M Series A (Prosus) |
| Arabic.AI / Tarjama | Enterprise agents | $$$ | $15M Series A 2025 |
§2.5 Why Araby.ai is beatable
Araby.ai looks like the giant in this space on paper. The firmographic reality says otherwise.
| Signal | Araby.ai | Implication for us |
|---|---|---|
| Headcount | 11–20, with 1 engineer listed | Can't out-ship us on narrow agents. Our 2 part-time engineers ≈ their engineering capacity. |
| Funding | Zero external — self-funded since 2022 | No war chest. We can out-spend them on ads if we need to. |
| ARR | ~$1M after ~4 years | Growth rate ~$250k/yr. Not explosive. PMF not dominant. |
| Revenue / employee | $86k | Mediocre for SaaS (healthy $200–300k+). Thin pricing power. |
| Product shape | Broad-shallow — all-in-one content suite (text, image, TTS, logo, translation) | We go narrow-deep. Each of our agents will be better than their equivalent feature. |
| Founder dependency | Single identified founder, thin management | Execution risk on their side, not ours. |
| Product gaps | No photo restoration. No podcast→shorts. No dialect-aware video. | Our agent #1 and #2 don't overlap their feature list at all. |
Araby captured the "generic Arabic content copilot" slot first but can't defend every vertical from it. Stretched across ~8 feature areas with 1 engineer. A focused 2-person team that picks one vertical and goes deeper will win that vertical every time. Our factory model explicitly targets verticals they can't build deeply.
We do not frame arabface.ai as "Araby alternative." We frame each agent as best-in-class at its narrow job; the portfolio emerges as a brand over time. Never compete on breadth — it's their only defensible story.
Gaps where MENA users currently workaround global tools
- Old Arab family photo restoration — Remini whitewashes Arab features (hundreds of 1★ reviews), MyHeritage is $299/yr + English-first + diaspora distrusts Western brands with grandfather's photo. No Arab-native player.
- Arabic dialect podcast→shorts — Opus Clip, Klap, Submagic handle MSA poorly, dialects not at all. Zero Arabic-native incumbent.
- Arabic dialect video generation (UGC/ads) — HeyGen/Synthesia lipsync fails on Arabic. Open.
- Arabic WhatsApp agent for solo entrepreneurs — WideBot/Arabot are enterprise-priced. No self-serve $29/mo tier.
- Arabic resume / job-apply for GCC market — low-end crowded; GCC ATS-tuned is open.
- Arabic Shopify / Salla / Zid support agent — Lucidya is enterprise-only; Salla alone has 50k+ merchants needing $29–99/mo tier.
Willingness to pay — proof in MENA
- Muslim Pro: 145M downloads, 25M MAU, $4.99/mo or $29.99/yr. Arab/Muslim audience pays recurring subs.
- Araby.ai: ~$1M ARR proves Arabic content creators pay even for a mediocre product.
- Tarteel: Quran memorization subs — deliberately chose subs over ads.
- Anghami: publicly listed, MENA paid subs work at scale.
Payment infrastructure (solved)
KSA mix: Apple Pay 36%, Mada 22%, cards 26%, STC Pay 12%, BNPL 5%. 85% of MENA consumers open to mobile wallet subs. Local payment support adds ~30% conversion lift (must-have by month 3).
Diaspora premium
Arab diaspora in US/UK/CA/DE pay USD for Arabic content tools (Muslim Pro, Anghami, Shahid, Quran apps confirm). The US-based co-founder is an asset — diaspora acquisition via Arabic-American community groups, mosques, Arabic Reddit, diaspora Facebook is a cleaner v1 channel than GCC enterprise.
§3 Why not a single-product bet?
Single-product risk: 2-person, 2–3 hrs/day team bets brand + time on one idea. Wrong = 60 days wasted.
Factory advantage:
- Portfolio absorbs misses. Agent #3 ships whether #1 hit or not.
- Shipping cadence is the marketing — every launch is a PR moment, a demo video, a TikTok share.
- Each agent teaches us what MENA actually pays for. Market research while earning, not before.
- Brand compounds. By agent #5, "arabface.ai has another tool" is a story pattern.
Caveat: factory has to start with 1–2 carefully chosen agents, not spray-and-pray. §4–5 is that choice.
§4 Agent #1 — arabface.ai/restore
Arab family photo restoration
What it does. Upload an old, damaged, faded, or black-and-white Arab family photo. AI restores, colorizes (culturally accurate — thobe colors, hijab, skin tones), enhances, optionally animates with respect (no Deep-Nostalgia kissing gimmicks). Bilingual UI.
Why this first
- Emotional pain = strongest sales force for a team without sales experience. Nobody needs convincing a torn photo of a late grandfather matters.
- Brand fits literally. "arab face" = restoring Arab faces. Validates the domain.
- Diaspora pays USD. 10M+ Yemeni + tens of millions of Syrian/Iraqi/Palestinian/Lebanese/Egyptian diaspora.
- Virality built-in. TikTok/IG before/afters are a proven format.
- No Arab-native competitor at quality. Remini whitewashes. MyHeritage $299/yr English-first. Fiverr $5 garbage. Arab-native slot is empty.
- Fast concept proof. Public before/afters + DM funnel — no B2B sales cycle.
Scoring: 32/35 — highest single score across 30+ candidates researched.
Stack & build
- Frontend
- Next.js 14 + Tailwind. RTL/LTR bilingual. Mobile-first.
- AI pipeline
- Flux.1 + CodeFormer/GFPGAN (face restore) + DDColor (colorization) + optional Live Portrait (respectful, off by default).
- Storage
- Cloudflare R2 — cheap, GDPR-friendly, auto-delete after 30d.
- Auth & pay
- Clerk + Stripe (USD first, Mada/STC Pay by month 3 via Moyasar).
- Support
- WhatsApp Business — diaspora converts on WhatsApp, not email.
- Build time
- 2–3 weeks part-time (both of us, 2–3 hrs/day).
Pricing
- Free: 1 photo, watermarked result
- Single: $9.99 per photo (clean HD, colorized, no watermark)
- Family Album: $29/mo — 10 photos + animation + optional voice narration
- Heritage: $99 one-time — 20 photos + booklet PDF
Concept-proof metric (day 60)
- Floor: 10 paying customers (any tier)
- Target: 50 paying customers + one reel >100k views
- Stretch: $1,000 MRR
Launch plan
- W1–2: MVP. Private alpha with 20 families from our networks.
- W3: Public launch. 5 before/after reels posted. Wamda + MENAbytes + Product Hunt.
- W4–6: 30 before/afters over 30 days (one per weekday). Co-founder → US-diaspora FB/Reddit. Abobakr → Arab Twitter + TikTok + WhatsApp.
- W7–8: Launch Family Album tier. Retention push.
- Privacy: diaspora is skeptical of Western data handling. Need explicit Arabic data-deletion policy + optional local processing mode.
- Moderation: deceased persons' photos need respectful tone. No kissing/smiling animation without explicit opt-in.
- Quality floor: if we ship whitewashed skin tones once, brand is dead. Manual review step for first 90 days (co-founder, US timezone).
§5 Agent #2 — arabface.ai/pod
Arabic podcast → shorts
What it does. Upload an Arabic podcast episode (MP3/MP4 or YouTube URL). AI picks the 5–10 best viral moments, generates dialect-correct captions, renders vertical shorts with Arabic-native typography, delivers ready-to-post clips.
Why second
- Infrastructure advantage. Abobakr's existing n8n-studio already has: Remotion long-form + shorts renderer, word-level Whisper captions, Gemini-2.5-pro QA chain, Vertex image pipeline, media-worker FastAPI, ElevenLabs integration. ~80% of this agent's backend is already built and tested. Competitors would need to build dialect-aware Arabic captions + Arabic typography from scratch.
- Clear B2B prospect list. Thmanyah (SRMG-acquired, 2.2M+ monthly listeners), Mics, Mstdfr Network, Kerning Cultures, AJ+ alumni pods. All MENA podcast studios with USD budgets. Co-founder's US network does warm email outreach.
- Market inflection. Saudi podcast market 33.6% CAGR → $2.56B by 2030. Thmanyah launched Radio Thmanyah Jan 2024 inviting creators.
- Real moat. Opus Clip, Klap, Submagic are English-first with MSA-only token Arabic. Dialect ASR + Arabic typography is genuinely hard to copy in a quarter.
Scoring: 25/30.
Stack & build
- Reuse
- Remotion renderer, Whisper word-captions, Gemini QA, media-worker, Vertex — all from n8n-studio.
- New work
- Arabic typography layer for vertical shorts (custom Arabic display fonts, kashida handling, RTL composition); viral-moment ranker (Gemini prompt tuned to Arabic rhetorical rhythms); customer Next.js UI; billing + team accounts.
- Build time
- 2 weeks part-time (weeks 4–5 of the 60-day plan).
Pricing
- Starter: $29/mo — 5 episodes/mo, 5 shorts each
- Pro: $79/mo — 20 episodes/mo, 10 shorts each, priority render queue
- Studio: $249/mo — unlimited, team seats, white-label export
Concept-proof metric (day 60)
- Floor: 1 paying studio
- Target: 3–5 paying studios at Starter/Pro
- Stretch: 1 Studio-tier customer
Launch plan
- W4: Adapt n8n-studio rendering into customer-facing product. Private beta with 5 Arab podcasts.
- W5: Public launch + direct warm-email outreach to 20 named MENA studios. Co-founder closes on Zoom.
- W6–8: First paying studios. Iterate on dialect accuracy + caption design.
§6 Unit economics (best-effort, pre-build)
Agent #1 — Restore
- Per-photo paid: compute ~$0.10–0.25 (Flux + CodeFormer + DDColor, ~30s GPU) + Stripe $0.30+2.9%. At $9.99/photo: variable ~$0.70. Gross margin ~93%.
- Family Album ($29/mo, 10 photos): variable ~$3.64. Gross margin ~87%.
Agent #2 — Pod
- Whisper ASR ~$0.30/hr (or self-hosted $0) + Gemini QA + ranking ~$0.20 + Remotion render ~$0.50. At Starter $29/mo (5 eps): variable ~$5. Gross margin ~83%.
Fixed monthly (combined)
- Domain + DNS: $10
- Cloudflare R2 + Workers + Pages: $5–20
- Stripe, Clerk: $0 (free tiers to hundreds of users)
- Vertex / OpenAI / Replicate base: $50–100
- Optional Vercel / Fly hosting: +$20
- Tooling (Figma etc.): $20
Fixed: ~$100–200/mo.
Both agents combined at ~30 paying customers (avg $10–80) → ~$1,200 MRR → covers all fixed + leaves ~$800/mo profit split 50/50. That's the 60-day floor to target.
§7 Partnership split (proposal — open to discussion)
Roles
Abobakr (Yemen) — "Product + MENA"
- Product build (front/back, both agents)
- Arabic content, copy, support
- Arab Twitter / TikTok / WhatsApp distribution
- Quality control on dialect + cultural accuracy
- Bridges existing n8n-studio infrastructure into Agent #2
Co-founder (US) — "Capital + US + Sales"
- US-side engineering pair programming
- US sales: diaspora FB groups, Arab-American mosques/community, US–MENA enterprises
- B2B email outreach to Agent #2's named prospects
- Capital (first 6 months burn)
- US legal entity (LLC), Stripe account, IP holding
- US-timezone manual photo moderation (first 90 days)
Equity, capital, time, IP
- Equity
- Proposed 50/50. Open to weighted-for-capital / vesting / buyback if cleaner for the US entity.
- Capital
- Proposed ceiling: $10,000 total burn for first 6 months (GPU credits, Stripe, ads, tooling, legal, entity formation). Above ceiling → partner call before spending.
- Source
- Co-founder funds, treated as founder capital, not loan.
- Time
- Both: 2–3 hrs/day committed, flex to 4–6 on ship weeks. 2× weekly sync, shared kanban (Linear or GitHub Projects TBD).
- IP
- All work product owned by US LLC. Pre-existing n8n-studio code stays Abobakr's, relevant pieces licensed to the LLC royalty-free for Agent #2.
§8 Decisions needed this week
Before any code ships or spend commits.
| # | Decision | Default (proposed) | Owner |
|---|---|---|---|
| D1 | Agent #1 pick | Restore | both |
| D2 | Agent #2 pick | Pod | both |
| D3 | Brand commitment | Keep arabface.ai as-is | both |
| D4 | Legal entity | US LLC — Wyoming or Delaware | co-founder |
| D5 | Capital ceiling | $10k for 6 months | co-founder |
| D6 | Equity split | 50/50 | both |
| D7 | Time commitment | 2–3 hrs/day both, flex up | both |
| D8 | Stripe account holder | Co-founder (US) | co-founder |
| D9 | Go / pause / kill review date | Day 60 (2026-06-21) | both |
| D10 | Plan B if both agents miss | Fold into MASTER_PLAN, revisit Q3 | both |
- Is agent order right? Restore first = virality-led. Pod first = infrastructure-led. Defaults argue Restore first.
- Stress-test the arabface.ai name with 10 native Arabic speakers before launching? (Recommended — one week max.)
- Equity 50/50 — any reason for asymmetric split given capital commitment? (Most 2-founder companies land 50/50 or 55/45.)
- Kill criteria. Proposed: if day 60 gives us <10 paying customers across both agents AND no reel >50k views AND no warm B2B lead, we fold.
§9 Timeline
| Week | Milestone |
|---|---|
| –1 (now) | Partner alignment call. Lock D1–D10. LLC filing started. |
| 1 | Agent #1 scaffolding. Stripe US account live. Domain DNS confirmed. |
| 2 | Agent #1 MVP complete. Private alpha with 20 family photos from our networks. |
| 3 | Agent #1 public launch. First 5 viral-candidate reels. Wamda + MENAbytes outreach. |
| 4 | Agent #2 scaffolding begins. Agent #1 post-launch iteration. |
| 5 | Agent #2 launch to 20 named MENA podcasts (co-founder-led outreach). |
| 6 | Agent #1 Family Album tier live. First Agent #2 paying studios. |
| 7–8 | Measurement window. Iterate retention + dialect quality. |
| Day 60 | Go / double-down / pause / kill review. Written outcome memo. |
§10 After day 60 — scenarios
| Scenario | Action |
|---|---|
| Both hit (>20 combined paying, one viral reel, one paying studio) | Hire 1 part-time diaspora-Arabic support ($500–800/mo). Plan agent #3 (likely WhatsApp solo-entrepreneur or Arabic resume). |
| Only Restore hits | Double ad spend, push to 500 paying customers, defer Pod to Q3. |
| Only Pod hits | Focus agents #3–4 on adjacent creator-stack (thumbnail, dubbing, audiobook). |
| Both miss (<10 customers, no viral, no B2B) | Honest post-mortem. Fold arabface.ai into MASTER_PLAN as Q3 revisit. Keep domain parked. No sunk-cost spiral. |
§11 Mission track — Yemen Qat safety detector
Reserving a place for co-founder's original idea inside the roadmap, separate from the revenue-first agents.
What it is
A computer-vision detector to help Yemenis distinguish likely-pesticide-contaminated qat from clean qat, based on leaf/stem/color patterns and provenance. User submits a photo, gets a risk indicator with reasoning. Community-labeled dataset grows over time.
Why it matters
Yemen consumes the world's largest share of qat per capita. Pesticide poisoning is a real public-health problem with no consumer-facing tool. This is a meaningful problem almost no team is positioned to work on — our co-founder's lived context is a genuine moat.
Why it is NOT agent #1 (honest)
| Blocker | What it means |
|---|---|
| No training dataset exists | Labeled qat-safety image dataset doesn't exist. Building one = 3–6 months collection + validation, not 2–3 weeks. |
| Physical-harm liability on day 1 | False "safe" verdict = someone's health harmed. v1 would need medical-device-grade validation + insurance coverage we don't have. |
| Frames the brand narrowly | Launching arabface.ai with "qat toxicity" locks the brand as Yemen-humanitarian, not MENA AI platform. |
| TAM wrong for money-making thesis | Only Yemeni/Ethiopian/Somali/Djiboutian populations. Humanitarian mission ≠ 60-day concept proof. |
| US sales muscle doesn't reach the buyer | Distribution is rural Yemen, not US diaspora or MENA SMB. Partnership asymmetric on this one. |
| Doesn't validate the platform thesis | Even a success wouldn't tell us whether MENA pays for AI agents. Wrong experiment for 60-day proof. |
The proposed path — "Mission track" in parallel
Run this as a separate track to the revenue-first agents, funded by (not competing with) the money-making agents.
Phase 0 — Data foundation (months 1–3, ~2 hrs/week from Abobakr)
- Identify Yemen partnerships: medical faculties (Sana'a / Taiz / Aden), health orgs (MSF, WHO Yemen office), agriculture cooperatives, 2–3 qat vendor networks willing to label.
- Start a labeled-image collection pipeline (photos + lab-tested pesticide residue correlation where possible).
- Co-founder leads partnership outreach from US; Abobakr leads Yemen ground coordination.
Phase 1 — Closed pilot (months 4–6)
- v0 model trained on whatever dataset Phase 0 produces.
- Ship as a closed WhatsApp-based pilot with a consented vendor network (not public).
- Every answer carries "guidance, not medical verdict — always verify."
- Funded by early revenue from Restore + Pod (mission budget ~$200/mo).
Phase 2 — Public launch (months 7–9, conditional on data quality)
- Public release as
qat.arabface.aior a distinct mission domain. - Partnership with one credible Yemen health authority for endorsement + liability umbrella.
- Critical: by this point, the core brand is already built on Restore + Pod. Qat detector arrives as "arabface.ai's community initiative," not as the brand's face.
Phase 3 — Open it up
- Release the dataset and model weights as open-source under a humanitarian license.
- This is the move that earns regional press (Wamda, BBC Arabic, Al Jazeera). Bigger story than any single SaaS launch.
Include Qat in the roadmap as "Mission track, launches month 6–9 conditional on data foundation". Do not let it pull resources from agents #1–#4 before then. Reserve qat.arabface.ai or a separate mission domain now to protect the path.
If in the next 60 days a genuine dataset partner (Yemen medical university, health NGO) offers a labeled corpus on a platter — we re-evaluate timing. Not before.