Cycling through all six. Tap any point to stop.

Measured on six things

Sovereign AI gateway with PII anonymization for GCC enter...

Security

HTTPS enforced, X-Content-Type-Options, Referrer-Policy, SPF and DMARC all present. But no HSTS, CSP, frame protection or Permissions-Policy (6/12 probe points), and no SOC 2, ISO 27001 or pen-test report published — the PDPL and audit-trail claims are self-asserted.

Screenshots

1 / 2

LLM Fusion

About this app

LLM Fusion is an enterprise AI platform that acts as a sovereignty-first gateway to over 300 large language models. It's built for organizations in the GCC region that need to use cutting-edge AI while keeping sensitive data in-jurisdiction and compliant with local data protection laws. The platform's standout feature is a five-stage anonymization pipeline (DETECT → MASK → FORWARD → RECEIVE → RESTORE → AUDIT) that strips personally identifiable information (PII) from prompts before they reach any external model, then restores it in the response. This means client names, national IDs, addresses, and other sensitive details never leave the organization's control. All compute and storage are hosted on regional infrastructure serving the UAE and wider GCC, with zero outbound traffic to non-GCC regions unless explicitly configured. The platform supports Arabic names and Gulf address formats natively, and offers an opt-in GLiNER NER sidecar for enhanced Arabic detection. A unique 'Council...

Key features

1Five-stage PII anonymization pipeline (DETECT → MASK → FORWARD → RECEIVE → RESTORE → AUDIT) with Arabic name and Gulf address detection.
2Sovereignty-first gateway to 300+ LLMs with UAE/GCC regional hosting and zero outbound traffic outside the region unless configured.
3Council mode: three models answer in parallel, peer-rank each other anonymously, and a chairman synthesizes the final verdict.
4Hash-chained audit trail exportable in regulator format, plus org console with per-member model access and credit usage tracking.

Target customer profile

Drawn from the product itself, not from a survey.

Demographic

Law firms in the GCC handling sensitive client matters

Pain points

Client confidentiality risks when using AI; fear of data leaking to external model providers; regulatory compliance with PDPL

Primary needs

Secure AI chat that anonymizes client names, matters, and amounts before sending to models; audit trail for regulatory review

Demographic

Fintech startups in the UAE and Saudi Arabia

Pain points

Balancing speed of AI adoption with strict data protection regulations; customer data exposure in AI prompts; lack of governed API paths

Primary needs

PDPL-ready audit trail; customer data masking before every model call; scalable API access to multiple LLMs

Demographic

Accounting and advisory firms with Gulf-based clients

Pain points

Staff using AI tools that may leak client identities; no control over where data is processed; Arabic name detection gaps

Primary needs

De-identification of client names, matters, and contact details in flight; end-to-end logging; Arabic-aware PII detection

Scout Review

Written by AI from measured evidence, scored out of 100.

LLM Fusion is a sovereignty-first AI gateway aimed squarely at GCC professional firms that want frontier models without leaking client data. The pitch is specific and the mechanics are real: a five-stage DETECT→MASK→FORWARD→RECEIVE→RESTORE→AUDIT pipeline, Arabic-native PII detection, hash-chained audit trail, and a Council mode where three models peer-review each other anonymously. The site is fast (Lighthouse 95 mobile) and the docs are unusually honest for a young product. The gaps are the ones that matter for a compliance sale: no SOC 2 or ISO certificate, thin HTTP security headers, and pricing that hides per-model cost behind a credit abstraction. It's a credible wedge against Azure OpenAI and Vertex for firms that can't route data abroad — provided the buyer accepts self-attested compliance.

Measure by measure

Usability

Docs are a real start-to-finish guide — org setup, Direct/Auto/Council modes, invites, credits, API keys — and $5 trial credits with no card lower the entry bar. Pricing stays vague on per-model cost, and the two-tier credit model hides real spend until you're inside.

Design

Clean dark enterprise look with a coherent pipeline diagram and vertical-specific cards that show real masking examples. Polished but template-shaped: gradient hero, stat trio, no customer logos or product screenshots to prove the UI.

Speed

Lighthouse mobile 95/100 (LCP 1.7s, CLS 0, TBT 5ms) and desktop 96/100. The marketing site is genuinely snappy with no layout shift — among the faster pages in this category.

Security

Accessibility

Lighthouse 94/100 with three real failures: contrast ratio, missing main landmark, and visible-text/accessible-name mismatch. No WCAG statement published, but bilingual EN/AR content and responsive layout help regional reach.

Growth

Sharp wedge: sovereignty-first gateway with in-flight PII masking, Arabic-native detection, and hash-chained audit — none of the listed alternatives (OpenAI, Azure, Anthropic, Vertex) ship that combo. Named verticals give it a real beachhead; the ceiling is how many GCC firms actually buy.

Improvement areas

No SOC 2, ISO 27001 or published pen-test report — compliance claims are self-attested, which is a hard sell to regulated buyers.
Missing HSTS, CSP, X-Frame-Options and Permissions-Policy; only 6/12 HTTP security header checklist points.
Pricing hides per-model cost behind a credit abstraction — '$1 = 100 credits' doesn't tell buyers what a call actually costs.
Accessibility gaps: failing contrast, no main landmark, and accessible-name mismatches despite a 94 Lighthouse score.

Design & Usability

The marketing site is competent enterprise SaaS: dark theme, a numbered DETECT→AUDIT pipeline diagram, and vertical cards for law, accounting, fintech, property and logistics that each show a concrete masking example. It's coherent but template-shaped — no product screenshots, no customer logos, no demo video, so you're trusting the copy. Usability is stronger than the design: the docs page is a real start-to-finish guide covering org creation, the Direct/Auto/Council model picker, team invites, credit top-ups and API keys, and the $5 no-card trial lowers the entry bar. The weak spot is pricing — '$1 = 100 credits' tells you nothing about what a GPT-class call actually costs, so real spend stays invisible until you're inside the console.

Speed & Security

Speed is the standout measured result: Lighthouse mobile 95/100 with LCP 1.7s, CLS 0 and TBT 5ms, desktop 96/100 — the site loads fast and doesn't shift. Security is the honest weak point for a product whose entire pitch is sovereignty. The HTTP probe found HTTPS enforced, X-Content-Type-Options, Referrer-Policy, SPF and DMARC, but no HSTS, no CSP, no frame protection and no Permissions-Policy — 6 of 12 checklist points. More importantly, the PDPL alignment, hash-chained audit trail and 'zero phone home' claims are all self-asserted: no SOC 2, ISO 27001, pen-test report or bug bounty surfaced. The docs are admirably candid that PII detection is probabilistic and outputs should be treated as drafts, which is the right caveat — but it isn't a certification.

Accessibility & Growth

Accessibility scores 94/100 on Lighthouse but fails three real audits: insufficient color contrast, no main landmark, and visible-text labels that don't match accessible names. No WCAG conformance statement is published, though bilingual EN/AR content and a responsive layout are genuine regional strengths. Growth is where the idea earns its keep: the wedge — in-flight PII masking with native Arabic name and Gulf address detection, GCC-jurisdiction hosting, and a hash-chained audit trail — is something none of the four listed alternatives (OpenAI API, Azure OpenAI, Anthropic, Vertex AI) ship out of the box. The Council peer-review mode is a real differentiator too. The limit is market size: GCC law firms, accounting practices and fintechs under PDPL is a specific, underserved segment, but a narrow one, and the domain is roughly four months old with only a LinkedIn presence — early stage, not a verdict on the idea.

Conclusion

If you're a GCC law firm, accounting practice or fintech that has been told 'no AI' by compliance, LLM Fusion is the most specific answer in this category — the Arabic PII handling and audit trail are not things the hyperscalers will build for you. Go in with eyes open: treat the PDPL claims as vendor assertions until you see a pen-test or SOC 2, ask for the audit-trail export format before signing, and pressure-test the masking on real Arabic names and Gulf addresses during the $5 trial. The idea is strong; the compliance paperwork is the work left to do.

Suggestions for improvement

Publish a SOC 2 Type II or ISO 27001 roadmap with dates — a sovereignty pitch without a certificate is a hard sell to regulated buyers.
Add HSTS, a Content-Security-Policy, X-Frame-Options and Permissions-Policy; 6/12 header points undercuts the security narrative.
Show per-model credit costs on the pricing page so '$1 = 100 credits' becomes a real number, not a mystery.
Ship a product screenshot or 60-second demo of the chat and audit-trail UI — the site sells the pipeline, not the product.

How we score

Alternatives comparison

Named competitors, point by point. Nobody paid to appear here or to be left out.

	LLM Fusion	OpenAI API	Azure OpenAI Service	Anthropic Claude API	Google Vertex AI
Built-in PII anonymization before model calls	Yes — five-stage DETECT/MASK/FORWARD/RECEIVE/RESTORE/AUDIT pipeline with Arabic support	No — developer must implement masking themselves	No native Arabic PII masking pipeline	No built-in anonymization	No native PII masking for prompts; DLP is a separate product
GCC / UAE data residency by default	Yes — UAE & GCC regional hosting, zero outbound outside region unless configured	No — global processing, no GCC residency guarantee	Partial — depends on regional deployment configuration	No — no GCC data residency	Partial — depends on region selection
Multi-model access in one gateway	Yes — 300+ providers via OpenRouter-style routing	No — OpenAI models only	No — OpenAI models via Azure only	No — Claude models only	Partial — Gemini plus select third-party models
Regulator-ready audit trail	Yes — hash-chained audit trail, DPO-ready export format	No — usage logs only, not compliance-formatted	Partial — Azure Monitor logging, not PDPL-formatted	No — no compliance audit export	Partial — Cloud Audit Logs, not PDPL-specific
Published security certifications (SOC 2 / ISO 27001)	Not found — PDPL alignment claimed but no certificate published	Yes — SOC 2 Type II, ISO 27001	Yes — extensive Microsoft compliance portfolio	Yes — SOC 2 Type II, ISO 27001, ISO 42001	Yes — SOC 1/2/3, ISO 27001, ISO 27017/27018
Pricing model	Credit-based: $1 = 100 credits, $5 free trial, org plans from $50/mo	Per-token pricing, pay-as-you-go	Per-token, enterprise contracts	Per-token, tiered by model	Per-token plus platform fees

OpenAI API

Direct access to OpenAI's models, but lacks built-in PII anonymization and GCC-sovereign hosting. Users must implement their own data protection measures.

Azure OpenAI Service

Enterprise-grade AI with Microsoft's compliance and security, but data may be processed outside the GCC unless specific regional deployments are configured. No native Arabic PII masking pipeline.

Anthropic Claude API

Access to Claude models with strong safety features, but no built-in anonymization or GCC data residency. Requires additional tooling for PDPL compliance.

Google Vertex AI

Comprehensive ML platform with generative AI models, but data sovereignty depends on region selection and may not meet strict GCC PDPL requirements without additional safeguards.

Comparing options? See LLM Fusion alternatives, scored side by side

View all Security