Three depth-planes = where your plaintext is actually processed. CloakAPI is the only engine that lives entirely on your side of the wire — on your device and, via the self-hosted proxy, inside your own infrastructure — so a vendor never sees plaintext. Everyone else processes it on your infrastructure or in their cloud.
Honesty first
Where CloakAPI loses.
A comparison you can trust has to start with the cells that don’t flatter us. Every card below is a technical loss: a measured number, a capability our code does not have, or an openness property a competitor genuinely holds. If one matters most for your situation, that system may be the better pick.
Business stage — not a moat
A moat is something cryptographic, mathematical or code-anchored — provable, testable, defensible. The four items below are none of that: any competitor with money can acquire them without changing a line of technology, and so can we.
Certifications acquirable
No SOC 2, ISO 27001 or HIPAA attestation yet — on the roadmap. Until then, the client-side architecture and signed receipts let you audit behaviour directly rather than paperwork.
Funding acquirable
Bootstrapped and solo-operated, versus Securiti’s $1.725B Veeam exit, BigID’s ~$320M raised, and three hyperscalers. Capital buys reach; it does not change what an engine can prove.
Brand acquirable
A new name next to incumbents nobody gets fired for buying. We ask to be judged on measured numbers and third-party-verifiable receipts instead.
Enterprise sales acquirable
No sales team, no procurement concierge, no quarterly business reviews — today. A hiring plan away for anyone, including us; never a property of the technology.
Measured, not marketed
The numbers we can defend.
Same corpus, same scorer, out-of-the-box settings: 320 synthetic texts,
40 languages (17 non-Latin scripts, 12 scripts), 1 340 annotated entities, recall and
precision scored by span overlap. Run 2026-07-25 against the frozen v1.2.1 engine
(ruleset 1.4.29 / WASM 00a4c7c7) — the same build shipped in the products.
CloakAPI figures below are the shipped default (engine + on-device NER). Corpus and runner are
prepared for public release (publication pending). Systems we haven’t run yet are marked
pending — not assumed.
The full ledger
39 factors. 20 systems.
Every cell labelled.
The 3D field above is a projection of this ledger’s MOAT rows only. 33 factors are provable properties of the technology; 6 are STAGE facts anyone can acquire.
Sources for vendor-stated cells are numbered — full list in the methodology. Cells marked “—” mean the capability is out of the system’s scope or we found no reliable public information. Vendor capabilities change; found an error? Tell us and we’ll fix it.
Four readers, four maps
What you should weigh.
The same twenty systems rank differently depending on who is asking. Each card re-projects the field onto the three axes that audience tends to care about most — including the axes where we are not the leader.
Methodology & sources
How this page was built.
What MEASURED means
Numbers labelled MEASURED come from our own benchmark run on
2026-07-25: a 320-text, 40-language synthetic corpus (1 340 gold entities; every text and every
identity fully synthetic — checksum-valid but fictitious IDs; place names verified against Wikidata;
the corpus is pinned by a sha-256 manifest), scored with one shared span-overlap scorer for every
system. Coverage counts (PII types, IBAN countries, checksum families, date-grammar languages) are
counted directly in our shipping engine's data files. CloakAPI numbers are pinned to the frozen
v1.2.1 engine that ships in the products (ruleset 1.4.29 / WASM 00a4c7c7) — not a
pre-release build — and are quoted for the shipped default configuration (engine + on-device NER).
Where a baseline beats us, the cell says so.
What STATED means
Cells labelled STATED repeat a vendor’s own published claim, with a numbered citation and the date we accessed it. We have not independently verified them, and a vendor’s own benchmark conditions usually differ from ours. Certifications (SOC 2, ISO 27001, HIPAA) are always vendor-stated — including the fact that CloakAPI holds none of them yet.
What PENDING means
An empty promise is worse than an empty cell. PENDING marks measurements we intend to run on our public benchmark but have not completed — e.g. the OpenAI Privacy Filter baseline is downloaded and wired but not yet scored at publication quality.
What MOAT vs BUSINESS-STAGE means
Every factor in the ledger is tagged MOAT or STAGE. A moat factor is cryptographic, mathematical, code-anchored or measured: a deterministic 100%-on-valid-input guarantee, client-side execution the SDK lets you verify, a signed receipt a third party can check, reversibility and conversation-scoped determinism, no-collision surrogates, counted coverage (115 types, 109 IBAN countries, 324 date-grammar languages, 23 checksum families) — properties that are provable and defensible, for us or against us. A stage factor (certifications, funding, brand, pricing posture, maintenance staffing) is a business fact any competitor with money can acquire without changing the technology — and so can we. Stage factors appear in the ledger for honesty but are never counted as technical wins or losses and never drive the 3D map: only moat axes position the bodies. Framing “no SOC 2 / bootstrapped / small brand” as a technical disadvantage would be as misleading as hiding it; we do neither.
What this page is not
Not legal or compliance advice — GDPR/HIPAA fit depends on your processing context, and no tool alone makes you compliant. Not a claim that CloakAPI is best for every job: the losses section is part of the spec for this page, not an afterthought. Vendor facts drift; the accessed date is on every citation.