Fleet-wide AI privacy, one executive view.
Argos runs as a transparent proxy across every device. The Executive Portal aggregates anonymized metadata from every agent — giving you org-wide visibility with zero raw data exposure. Try it: edit the prompt and watch what the model would actually receive.
The portal runs entirely client-side. Everything you see is realistic demo data.
One roundtrip: tokenize, scale, forward, rehydrate.
The proxy rewrites sensitive bytes before they leave the box and restores them in the response, so your users never notice. Here is the whole trip.
Tokenize
Every detected PII span is swapped for a deterministic [CAT:SUBCAT:NNNN] placeholder before the request leaves the box.
Scale
The Prism Method multiplies revenue, headcount and other sensitive magnitudes by a cryptographic factor. Ratios preserved, secrets hidden.
Forward
The tokenized payload travels to OpenAI, Anthropic, Gemini or 13 more. Streaming SSE plus gzip and brotli are handled statefully per vendor.
Rehydrate
The response stream is decompressed, placeholders restored, magnitudes un-scaled. The user sees a coherent answer in plain English.
Org-wide visibility. Zero raw data.
The Executive Portal aggregates anonymized metadata from every reporting agent. Seven views cover everything from fleet overview to per-device drill-downs, department breakdowns, and board-ready compliance reports.
Regex catches the obvious. K-anonymity catches the rest.
Detection runs end-to-end in under 50ms. Nothing is sent off-device for analysis: every layer ships with the agent.
Regex
SSN, credit card, IBAN, ICD-10, API keys, JWT, private keys, DB connection strings, emails. Deterministic, sub-millisecond.
Semantic NER
ML named-entity recognition for personal names, medical conditions, minors and special categories that escape pattern matching.
The combination attack. Where regex stops, Sweeney begins.
"John Smith" is safe. "94304" is safe. "1987-03-12" is safe. Put them in one prompt and 87% of the US population is uniquely identifiable. Argos models re-identification risk with Sweeney k-anonymity and HIPAA Safe Harbor, and blocks combinations regex will never see.
findings: 4 spans · layer_3_k: 1.0
combo: hipaa_safe_harbor · sweeney_quasi_identifier
verdict: CRITICAL → action: BLOCK
Metadata only
The central reporter sends counts, scores and labels, never raw findings. The PII string never crosses the network.
Response-side scan
Models hallucinate emails and paraphrase names. Argos independently rescans the model's output and logs it as a separate model-generated PII event.
What ships today.
Every endpoint treated like a target.
v2.2 ships with the OWASP API Top-10 mitigations the average self-hosted security tool quietly ignores.
Adaptive rate limiting
Per-IP and per-identity controls on every API surface. Ingest, dashboard, login and admin each enforce separate thresholds automatically.
Strict input validation
Every request is gated by a strict schema that rejects unexpected fields. Errors never echo raw input back to the caller.
Hashed token storage
Authentication tokens are stored only as cryptographic hashes and verified using constant-time comparison to prevent timing attacks.
Hardened session management
Auth cookies are inaccessible to client-side scripts, auto-secured on HTTPS, and protected against clickjacking with strict framing policies.
Decompression-bomb caps
50 MB cumulative output ceiling per stream. Per-format guards on PDF, DOCX, XLSX and PPTX uploads.
Zero-knowledge reporter
The findings list, the only part of the scan that holds matched PII, is dropped before the payload is queued for the central server.
Run it now, in this browser.
No sign-up, no install, no backend. The Executive Portal works entirely offline against realistic demo data; every filter, chart and export is live.