SAJ
The Sovereign Perception Stack.
Sees. Hears. Speaks. Senses.
Saj — pronounced “Say.” The perception stack that sees, hears, speaks, and senses — on device, Arabic-first, sovereign by design.
While you opened this deck,
Saj senses are streaming.
Saj Speak heard intent in a Gulf-Arabic dialect. Saj See lifted text off the screen on-device. Saj Link wrapped the channel in an end-to-end-encrypted tunnel over local infrastructure. Saj Sense unified all four streams into one multimodal token stream — no cloud round-trips, no jurisdiction leaks.
This is what production-ready engineering looks like, paired with pre-revenue honesty about the channel work still ahead. First paid customer target Q3 2026.
AI perception is cloud-locked, fragmented, and Anglo-only.
MENA and on-device users get nothing.
Cloud-only AI leaks data
Every API call to OpenAI, Anthropic, Google, ElevenLabs ships raw user audio / images / text off-device into US/EU jurisdictions. Sovereign-tier customers (banks, govs, defence, healthcare) cannot adopt at scale — data residency + EU AI Act + GCC PDPL all bite.
No Picovoice for multimodal
Picovoice proved an on-device voice SDK works ($1.5M ARR / 16-person team). But voice is one sense. Voice + vision + sensor + comms are still four separate vendors, four SDKs, four billing relationships. Developers stitch glue forever.
Arabic is a second-class citizen
Every major perception SDK was built English-first then back-translated. Arabic dialects, RTL layout, mixed-script OCR, GCC sovereign hosting — treated as “optional locale”. 470M Arabic speakers + Vision 2030 + Bahrain Vision 2030 demand are not addressed by any incumbent.
Every incumbent treats these as features to add. None can rewrite the perception architecture beneath their cloud-only installed base.
Four senses. One engine. One token stream.
Whatever you need it to Say.
One multimodal token stream — voice, vision, text, sensor
Saj Sense fuses Speak (voice) + See (vision) + Link (comms) + signal-layer telemetry into a single unified token stream addressable by any downstream LLM or rule engine. The contract every Saj product targets.
Frozen-codec RVQ + watermarked perception tokens
Cross-modal residual vector quantisation, domain-adaptive tokenisation, neural watermarking (PROV-038/-080). The engine inside every Saj product — and IP-licensable to OEMs under EU AI Act Article 50 (Digimarc precedent).
Each layer is independently testable, replaceable, and crypto-bounded. None of the incumbents can ship Layer 1 without re-architecting their cloud-only stack.
Two markets. One stack.
Live in production today.
17 engines, 4 marketing sites, 1 app — in production
- sajlink.com / app.sajlink.com (iOS TestFlight live)
- sajspeak.com (voice synthesis · ASR / TTS)
- sajsee.com (visual intelligence · OCR + scene)
- sajsense.com (multimodal signal fusion)
- Saj Codec engine — 17 codec / perception engines
- Cross-platform SDK · iOS + Android + Web + Linux
There is no on-device perception stack for MENA. We built it.
Arabic-first across all 4 products — voice (Gulf dialect ASR/TTS), vision (OCR-ar + RTL), comms (e2ee with KSA / UAE / Bahrain peering), sensor (PDPL-compliant on-device fusion). SGH WLL Bahrain entity active; Saudi Vision 2030 + Bahrain Vision 2030 aligned. A A$1.1B Activate envelope addressable in-region with zero category incumbent.
Saj Codec ships neural watermarking (PROV-038/-080) + cross-modal RVQ (PROV-030) + domain-adaptive tokenisation (PROV-037). EU AI Act Article 50 mandates content provenance from Aug 2026 — Digimarc / Truepic precedent shows A$0.50–A$5.00 / device / yr licensing economics. Future opportunity, not current revenue.
Picovoice can ship a multimodal feature.
They cannot ship a multimodal architecture.
Defensibility score · weighted by patent claim breadth + reproduction cost + regulatory tailwind
“EU AI Act Article 50 mandates content provenance from Aug 2026. Digimarc's market cap on a single watermarking patent class is > A$500M.”
Saj is the only player with all four senses + on-device + Arabic-first + 4 crown-jewel patents.
A$45B perception. A$5B sovereign-native.
Three tailwinds converging at once.
Four developer tiers + enterprise + OEM.
Designed to undercut Picovoice on day one.
- · 1,000 frames / mo
- · 1 engine (pick one)
- · Watermark "Saj Free"
- · Community support
- · 25,000 frames / mo
- · All 17 engines
- · Arabic-native models
- · Discord support
- · 250,000 frames / mo
- · All 17 engines + custom
- · 5 seats + Slack support
- · PDPL-compliant region
- · 5M frames / mo
- · SLA + dedicated AM
- · SOC 2 + sovereign DEK
- · OEM trial allowance
OEM Saj Codec licensing is a future revenue opportunity tied to EU AI Act Article 50 enforcement (Aug 2026 onward). Digimarc / Truepic precedent shown for reference; first OEM customer target Y2.
Picovoice pricing.
17 engines + Arabic + sovereign.
| Tier / Capability | SAJ | Picovoice | Hume AI | ElevenLabs | Roboflow | Mozn |
|---|---|---|---|---|---|---|
| Free dev tier | A$0 1K frames/mo |
Free 3 users |
paid only |
Free 10K char/mo |
Free 3 projects |
|
| Mid tier (US$ or A$/mo) | A$29 Developer |
US$499 Enterprise lite |
~US$55 Starter |
US$22 Starter |
US$249 Growth |
Custom |
| Team / Most-popular (A$/mo) | A$199 Team |
~A$1,500 Enterprise |
~A$330 Pro |
A$150 Creator |
~A$1,200 Business |
Custom |
| Enterprise (A$/mo) | A$5K+ Sovereign |
A$15K+ Custom |
A$10K+ | A$3K+ | A$8K+ | A$20K+ |
| Multimodal coverage (voice + vision + sensor + comms) | all 4 | Voice only | Voice + vision | Voice only | Vision only | Voice + text |
| Arabic-native models | Partial | Partial | ||||
| On-device inference | cloud | Edge tier | ||||
| Patent moat (filed / drafted) | 14 | ~3 | ~2 | 0 public | ~1 | |
| Free trial · no credit card |
At A$199 we ship 17 engines + Arabic-native + on-device — Picovoice doesn't offer Arabic at any tier; ElevenLabs is cloud-only; Roboflow has no voice.
We're not just undercutting Picovoice — we're the only player covering all four senses with sovereign-tier hosting.
Single-sense vendors, cloud-only stacks,
no Arabic-first perception player.
| Capability | SAJ | Picovoice | Hume AI | ElevenLabs | Roboflow | Mozn |
|---|---|---|---|---|---|---|
| Voice (ASR + TTS) on-device | ||||||
| Vision (OCR + scene understanding) on-device | ||||||
| Sovereign comms (e2ee Link layer) | ||||||
| Multimodal sensor fusion | ||||||
| Arabic-native models (Gulf dialect, RTL) | ||||||
| Neural codec watermarking (PROV-038/-080) | ||||||
| Unified multimodal token stream (SSP-TOKEN-001) | ||||||
| GCC sovereign hosting (PDPL / KSA / Bahrain) | ||||||
| Free dev tier with full engine access | 3 users | 10K char | 3 projects | |||
| EU AI Act Art 50 ready (content provenance) | ||||||
| 14-patent IP moat | ~3 | ~2 | 0 | ~1 | ||
| Cross-platform SDK (iOS + Android + Web + Linux) | Cloud SDK | Cloud SDK | Cloud SDK |
11 of 12 capabilities · categorically absent from the incumbent perception stack.
Four tailwinds converging.
One perception stack that meets all four.
EU AI Act Article 50 mandate
From Aug 2026 every AI-generated or AI-modified piece of content distributed in the EU must carry a machine-readable provenance mark. Neural watermarking (PROV-038 + PROV-080) is one of two compliant approaches. Digimarc's market cap on the precedent class: > A$500M.
Vision 2030 + Bahrain Vision 2030
Saudi Public Investment Fund + Bahrain Tamkeen actively funding sovereign-AI infrastructure. PDPL data-residency rules locked in. Western cloud-only AI vendors structurally excluded from gov + bank + healthcare procurement. Saj WLL Bahrain entity is active.
On-device AI is finally fast enough
Apple Neural Engine, Qualcomm Hexagon NPU, Intel AI Boost — on-device inference for sub-100ms multimodal perception is now viable on consumer hardware. Saj Link Web Mimi codec runs at 1.49× real-time on M1 in Chromium today (Lane G shipped 2026-05-15).
Picovoice + Digimarc proof points
Picovoice: A$1.5M ARR on voice-only on-device with a 16-person team. Digimarc: A$500M+ market cap on a single watermarking patent class. Hume AI: US$50M raised at US$219M post on emotional voice. Saj covers all four senses + Arabic + 14 patents.
Production-ready. Patent-armed. We've Said it; we've shipped it.
4 production subdomains + 1 product app.
sajlink · sajspeak · sajsee · sajsense + app.sajlink.com
A$1.80M founder-deployed Y0
iOS TestFlight Field Cohort live
Architecture nobody can retrofit.
8 filed AU PROV + 6 drafted ready
PROV-030 / PROV-037 / PROV-038 / PROV-080
+ SSP-TOKEN-001 crown jewel
Market clock is ticking.
Aug 2026 EU AI Act Art 50 mandate
Vision 2030 + Bahrain Vision 2030
Picovoice ($1.5M ARR / 16 ppl) precedent
Pre-revenue by design. First paid customer target Q3 2026.
Developer SDK. Enterprise sovereign. OEM strategic. Just Saj it.
1 npm install → 5,000 seats.
1 MENA gov logo → 1,000s of seats.
1 device maker → millions of devices.
EU AI Act Art 50
Solo + AI-augmented. 24 months.
14 patents armed.
Cross-Modal Predictive Compression
Joint audio + vision + text RVQ tokenisation; predictive coding across modalities. Filed AU 17 Mar 2026.
Domain-Adaptive Codec Tokenisation
Per-domain codec fine-tune (medical / legal / Gulf-Arabic) without re-training the base codec. Frozen-codec RVQ-bias.
Neural Codec Watermarking
Latent-perturbation + frozen-codec logit-bias signature schemes. PROV-080 Y2 empirically validated 2026-05-15.
Unified Multimodal Token Stream
Single addressable token stream across all 4 senses (Saj Sense protocol). The contract every Saj product targets.
14 patents armed · 4 crown jewels above · 10 supporting filings across audio + vision + sensor + comms.
A$1.80M already deployed.
A$10M to compound it.
Match the capital that's already inside.
We've Said our piece. Your move.
A$350M–A$1.2B to Apple / Samsung / Google / Qualcomm at sovereign-AI + EU AI Act compliance + Arabic-native premium.
PIF / Mubadala / Saudi Aramco-style strategic acquisition for Vision 2030 sovereign-tech IP at Base A$29M Y3 ARR · Bull A$52M Y3 ARR.
24% engineering / 18% MENA sales / 15% ML research / 13% dev marketing / 13% buffer / 9% security / 8% legal & IP.
Just Saj it.