Purfect Labs · Product Demo

Shield

Where the rivers meet.

A 25-minute unedited walkthrough of the middleware layer that strips secrets on the way out and restores them on the way back — recorded against a live enterprise stack, not a mockup.

Shield is a Purfect Labs product — our umbrella for work across AI security, blockchain, cloud infrastructure, applied AI and MLOps.

213
Providers routed
3.23B
Input tokens
100%
Traffic scanned
71.1K
Tool calls
25:13
Unedited runtime

The 40-second cut

Narrated · start here
Short Shield in 40 seconds 00:41

The full walkthrough

25:13 · unedited, live stack
Now Cold start — the command palette 00:00 / 25:13

Jump to a chapter

Click any card to seek

The proof, at 14:35

Five signal types, one prompt each

Shield's demo corpus runs one clean signal per line. Below is exactly what the model receives — transcribed from the sanitized panel in the recording.

SSNpii:ssn
Original · stored, never sent
Hi, this is Marcus Delgado — I'm filing a correction and my social is 219-09-1234. Can you draft the cover letter?
Sanitized · what the model sees
Hi, this is {{REDACTED_PII_PERSON_91EE9AB2_288}} — I'm filing a correction and my social is {{REDACTED_PII_SSN_4489AD8C_907}}. Can you draft the cover letter?
Emailpii:email
Sanitized · what the model sees
Please send the summary to {{REDACTED_PII_PERSON_6D77F4F7_116}} at {{REDACTED_PII_EMAIL_6D362D29_422}} when it's ready.
Credit cardpii.credit_card · Visa test number, Luhn-valid
Sanitized · what the model sees
Charge {{REDACTED_PII_PERSON_095D943A_226}}'s annual plan to {{REDACTED_PII_CREDIT_CARD_9F876D02_912}}, exp 08/27, cvv 123.
Bank routingpii.routing_number · ABA-checksum-valid
Sanitized · what the model sees
Set up the ACH pull for {{REDACTED_PII_PERSON_91EE9AB2_288}} from routing {{REDACTED_PII_SSN_E2F31A19_566}}, checking account.
Date of birthpii:dob · context-gated
Sanitized · what the model sees
Enrolling a patient — {{REDACTED_PII_PERSON_NAME_LABELED_A2322B26_886}}, please summarize eligibility.
Placeholders are stable and reversible — the same person resolves to the same label across prompts (91EE9AB2_288 carries the SSN line and the ACH line), so the model keeps the thread while Shield restores real values on the response path. 325 redactions across 67 distinct values in this run.
Context-gated, not pattern-blind: a date following "DOB", "born" or "date of birth" gets caught, while an ordinary date like the deploy ran on 2026-08-19 is left alone. Shield doesn't redact every timestamp in a code request.
All demo values synthetic — SSA-valid-but-fake SSN areas, Luhn-valid test cards

Why Shield

The gap nothing else covers

Your DLP never saw this coming

Data-loss tooling was built for email, uploads and USB drives. An agent calling a model over HTTPS from a developer's laptop looks like ordinary traffic. The payload is a prompt, and nobody is reading prompts.

Policy is not a control

"Don't paste customer data into AI tools" is a sentence in a handbook. It is not enforcement. The only place a rule like that can actually be applied is on the wire, in the moment, before the request leaves.

Blocking the tools loses

Ban the assistant and your team uses it on a personal laptop instead — same leak, now with no logs at all. Shield assumes the tools are staying and makes them safe, rather than pretending adoption can be reversed.

Redaction that doesn't break the work

Stripping data is easy; keeping the answer useful is the hard part. Stable, reversible labels mean the model still reasons about the same entity, and your side gets the real values back. Security that costs nothing in output quality.

Evidence, not assurances

When an auditor asks what your AI tools have touched, "we have a policy" is not an answer. A hash-chained local audit log turns every redaction into tamper-evident evidence with categories, actors and timestamps.

It runs on your machine

Shield is not a cloud service standing between you and your provider. One binary, one config, local gateway. A single license check at activation and no other outbound call. Nothing to trust us with.

Two ways to run it

Subscribe, or own the source
The app · self-serve

One binary. One config.

Shield runs on your machine, not ours. Gateway, CLI, dashboard and audit log all in one binary. No hosted service, no data pipeline.
Shield Core$49.99/mo
Shield Pro$79.99/mo
Shield Max$99.99/mo
  • Plugins auto-install for the CLIs your team already uses
  • Desktop coverage with per-app health checks
  • Transparent redaction — the developer never notices
  • Annual pricing available
The platform · source-delivered

You own it. Forever.

Scoped engagements that end with you holding the source. No subscription, no license server, no kill switch. Roughly a week per tier.
FoundationScoped
ComplianceScoped
CustomScoped
  • Gateway on your infra, local or remote middleware
  • Threat-model workshop and 30+ eval cases, all green
  • Audit sidecar, dashboard, SOC 2 / HIPAA export
  • Semantic detection, entity linkage, federation, GPU orchestration
  • Runbook and full Go source at handoff

Work with us

Partnership · reseller · integration

Shield is a product and a capability. If you serve teams that are adopting AI tools faster than they can secure them, there is a version of this that works for you. Every route below reaches the engineers who build it — pricing is scoped per engagement.

01 · Partnership

Co-sell and joint delivery

Bring Shield to your accounts and deliver alongside us. We handle the security architecture; you keep the client relationship you already earned.

Start a conversation →
02 · Reseller

Resell under your agreement

For MSPs, MSSPs, consultancies and compliance advisors with an existing book. License Shield through your own paper and your own margin.

Request reseller terms →
03 · Agent interception

Put a layer in front of your agents

You have autonomous systems calling models and no visibility into what they send. We intercept that traffic at the gateway, classify it, and give you a control point and an audit trail.

Scope an interception →
04 · Integration & OEM

Shield inside your product

Go interfaces as extension points, not a plugin registry. Embed the redaction layer in your own platform, or wire our filters into your build pipeline and CI eval gates.

Discuss integration →
05 · Consultation

Threat modeling and readiness

Short, scoped engagements: map where your AI tools actually send data, model the exposure, write the filter packs for your domain, and prove the coverage with eval cases.

Book a session →
06 · Shield as a service

Managed, on your infrastructure

You want the outcome without running it. We deploy and operate the gateway inside your environment — your keys, your bucket, your boundary — and hand over the evidence.

Ask about managed →

Who builds this

Purfect Labs Inc.
Shawn Meredith
Founder & Lead Architect

Low-latency systems engineer and DevOps pioneer. Author of the Hermes-v4 daemon chassis, with deep work in sandboxed execution environments, telemetry pipelines and memory vectors. Shield's gateway, plugin supervisor and audit chain come out of that lineage — infrastructure built to run unattended and be inspected after the fact.

Hermes daemonDocker sandboxingTelemetry & logsGo
LinkedIn
Miss Purfect
Principal AI Systems Engineer

Systems architect working in large language model pipelines, deterministic scaffolding and vertical compliance kits. Leads intake compiler specs and prompt orchestration — the side of Shield that decides what counts as sensitive in a given domain, and proves the filters hold with eval cases before anything ships.

LLM pipelinesCompliance guardrailsFilter packs & evalsPrompt architecture
LinkedIn
We are a small team that ships production systems directly — no junior handoffs, no account managers. Shield exists because we kept watching capable people hand real credentials and real customer data to tools that were never designed to hold them. The full team, product detail and current pricing live at purfectlabs.com — this page is the demo that sits behind it.

Purfect Labs is the umbrella, not the product

We founded Purfect Labs Inc. in Utah in 2026 as a home for several lines of work rather than a single app. Shield is the one that got far enough to sell, but the company was built to carry more than one bet — and the disciplines feed each other. The audit chain behind Shield's evidence trail is the same thinking that makes a ledger tamper-evident; the gateway that intercepts an agent is the same infrastructure work as securing a cloud deployment.

AI security
Shield: redaction gateways, filter packs, agent interception, audit evidence.
Blockchain
Signed ledgers, content-addressed structures and proof-carrying data — applied as primitives inside products, not as a separate coin.
Cloud security & builds
Sandboxed execution, provisioned fleets, telemetry pipelines and deployments that survive being left alone.
AI applications
End-to-end agent systems and developer tools taken from prototype to production.
MLOps
Model pipelines, eval harnesses and CI gates — the machinery that proves a system still behaves after you change it.
How we work
Scoped engagements, source delivered, principal engineers on every call.

Tell us what you're protecting

Goes straight to the engineers

Partnership, reseller terms, an integration, an agent-interception build, or a straight evaluation — pick a time or send us a note. You reach the engineers who build Shield, and you get back a production scope with architectural dependencies, not a brochure.

Principal engineers only
Every request goes to the people who build Shield. No juniors, no account managers.
Zero cloud risk
Contracts guarantee strict IP ownership and security boundary isolation.

One binary. One config. Tap.

Shield runs local on the desktop, or gets pushed to the whole fleet by MDM. Same middleware layer either way. Partnership, reseller, integration or a managed build — tell us what you're protecting.

Where the rivers meet