Skip to content
MULTINEX
AIEGES Shield

Multinex Shield

Stop sensitive data
before it reaches AI—or memory.

Shield is a local classifier and prompt guard. It inspects prompts, files, and tool calls on the device, redacts personal, health, and proprietary data, and keeps an audit record before anything is sent to an AI provider or saved to MemQ.

See Shield plansHow Shield protects MemQ

Desktop app for macOS and Windows, a browser extension, and a local MCP gateway. Model inference stays on the workstation.

PromptsChatGPT, Claude, Gemini, Copilot, IDE agents
Files + uploadsDocuments pasted or attached in the browser
MCP tool callsIncluding add_memory and save_context
Inspected on the device
AIEGES Shield

Multinex Shield

Local classifier + prompt guard

  1. 1
    Pattern scan365+ signatures: secrets, PII, PHI, injection, malware indicators
  2. 2
    Neural classifierQwen2.5-0.5B via ONNX Runtime, on-device, no cloud call
  3. 3
    Redact or blockReversible receipts; vault placeholders for credentials
AI providerSees the sanitized request
MemQStores the redacted, policy-checked memory
Audit recordDecision, labels, receipt id, no raw data
Shield runs as a desktop app, browser extension, and local MCP gateway. Model inference stays on the workstation.
How Shield sits in the path
What goes in
Prompts, files, and MCP tool calls from the tools your team already uses.
Shield
Pattern scan, on-device neural classifier, then redact or block with a receipt.
Where it goes
The AI provider sees the sanitized request. MemQ stores the redacted memory. The audit record keeps the decision, not the data.

What Shield catches

Three kinds of exposure. One local checkpoint.

Personal and health data

Names, identifiers, contact details, medical record numbers, and diagnosis codes are redacted before they leave the device.

  • PII and PHI pattern sets
  • Payment and account identifiers
  • Industry policies: legal, healthcare, financial

Secrets and credentials

API keys and tokens are moved into a local encrypted vault and replaced with a placeholder. The real value is substituted only when the call runs.

  • Sentinel KMS vault placeholders
  • Reversible redaction receipts
  • Developer and CI secret patterns

Prompt injection and unsafe instructions

Instructions hidden in pages, files, and tool output are classified before an agent acts on them.

  • 365+ threat signatures
  • Multi-language jailbreak detection
  • Block, warn, or review by policy

Shield + MemQ

The memory boundary for enterprise agents.

Shared memory is only safe if the write is safe. Agents call MemQ through MCP tools such as add_memory and save_context. With Shield in the path, every write is classified and redacted on the device first, so private data, health records, and proprietary details never become someone else’s recalled context.

  • Classified locallyThe local gateway scores the payload with pattern rules and the on-device neural model before routing it.
  • Redacted with a receiptSensitive values are replaced or vaulted. The receipt stays on the device; MemQ stores the sanitized memory.
  • Recorded for auditDecision, labels, and receipt id are logged and can be forwarded to Billing Manager for team review. Raw data is not.
  • Checked again by MemQMemQ’s gateway runs deterministic PII, PHI, and secret redaction on flagged writes and on promotion to shared knowledge.

Status. Shield’s local MCP gateway and MemQ’s server-side gate are available today. A native Shield-to-MemQ policy handshake, sharing policy bundles and receipts across both products, is planned.

Explore MemQ
add_memory · through ShieldIllustrative example

What the agent tried to save

Customer Jane Doe (MRN 00482913) asked about invoice INV-4471. Card ending 4242 is on file. Use sk-live-7f3a… to re-run the export.

What MemQ stores

Customer [REDACTED_PII] ([REDACTED_MRN]) asked about invoice INV-4471. A payment method is on file. Use <SHIELD:vault:export_key> to re-run the export.

Redacted
Name, medical record number, card digits
Vaulted
API key replaced with a Sentinel placeholder
Kept
Invoice reference and the task itself
Receipt
Reversible only on the device that redacted it

MemQ’s gateway applies its own deterministic PII, PHI, and secret redaction on flagged writes and on anything promoted to shared knowledge.

How it works

Install once. Work as usual.

1

Install Shield

Add the desktop app and load the browser extension. Policies sync from your Billing Manager account.

2

Choose a policy

Start with the data you most need to protect, then pick an industry pattern set or define your own.

3

Review and export

Shield redacts or blocks in place. Review decisions in the app and export evidence for compliance.

Built for compliance teams

Evidence you can hand to an auditor.

Everything Shield decides is recorded without recording the data it protected.

Local-first inference

Qwen2.5-0.5B runs through ONNX Runtime on the workstation. No prompt content is sent to a classification service.

Reversible redaction receipts

AES-SIV encrypted receipts let an authorized user restore a value on the device that redacted it.

Exportable audit log

Every decision carries a timestamp, labels, and receipt id. Export it for review or forward it to Billing Manager.

Team policy control plane

Publish policies, assign devices, and review audit exports for your organization from Billing Manager.

Shield documentation

Setup, policies, and rollout guidance.

Governance + policies

Data handling, access, and review.

MemQ privacy + access

Namespaces, sharing, and redaction gates.

Plans

Priced per person. Protection for every AI tool they use.

Every plan includes the desktop app, browser extension, local redaction, and exportable evidence.

1 user

Individual

$14.99 /mo

Single machine protection

Protect one user or practice owner from leaking sensitive data into AI tools.

  • Business AI data protection
  • Local sensitive-data redaction
  • Prompt and file screening
  • Exportable security evidence
Up to 10 users

Small Team

$129.90 /mo

$12.99/seat billed monthly

Protect the first team with browser, gateway, and policy coverage for business AI usage.

  • Shared team policies
  • Consistent browser protection
  • Extension and gateway coverage
  • Centralized security controls
Up to 25 users

Business

$299.75 /mo

$11.99/seat billed monthly

Protect departments and regulated practices with audit trails, analytics, and broader rollout coverage.

  • Audit-ready reporting
  • Advanced evidence export
  • Department-wide policy controls
  • Higher-volume protection
Up to 50 users

Organization

$499.50 /mo

$9.99/seat billed monthly

Protect larger AI organizations with centralized access governance and the best self-serve seat value.

  • Organization-wide protection
  • Larger team rollout support
  • Centralized access governance
  • Best self-serve seat value

Annual billing is available for every plan. Organizations over 50 users, air-gapped deployments, and policy-as-code rollouts are scoped with our team.

Start Shield in Billing ManagerPlan a team rollout

Before you install

A few useful answers.

Does Shield send my data to the cloud to classify it?

No. Pattern matching and the neural classifier run on the workstation through the desktop app, browser extension, and local gateway. Billing Manager receives license, policy, and audit metadata, not the content that was inspected.

How does Shield relate to MemQ?

Shield is the guard in front of the write. When an agent saves context through a Shield-protected MCP path, Shield classifies and redacts the payload before it reaches MemQ. MemQ then applies its own server-side redaction gate on flagged writes and on anything promoted to shared knowledge. Shield’s local gateway runs today; a native Shield-to-MemQ policy handshake is planned.

Can a redaction be reversed?

Redaction receipts are encrypted and reversible only on the device that produced them, by an authorized user. The AI provider and MemQ never receive the original value.

Which AI tools does the browser extension cover?

The extension hooks into the major assistant surfaces teams use in the browser, including ChatGPT, Claude, Gemini, and Copilot. Coverage for a specific tool is confirmed during rollout.

Ready to move forward?

Give your team AI tools without giving away the data.

Tell us which tools and data types matter most.
We’ll help scope a practical rollout.

Plan a team rollout

Or start with one seat.

Opens your email app. Please don’t include credentials or customer records.