SentraMind

Platform

A governed control plane for private AI security

SentraMind is designed for self-hosted and air-gap-capable operation. The current POC keeps deterministic detection and policy authoritative, records reviewable events, and makes every stand-in visible.

Current offer first. Future modules remain visibly separate.

Governed decision path

  1. 01 Simulated prompt input
  2. 02 Deterministic detector
  3. 03 Versioned policy resolution
  4. 04 Warning issued
  5. 05 Reviewable event record

Available for POC

Current product boundary

Know exactly what you can evaluate

A verified package does not make a whole capability production-ready, and a roadmap direction never borrows the visual language of current product.

Platform Core

Available for POC
Module
Core
Availability
Available for POC

Evaluate deterministic policy, role enforcement, immutable policy versions, and a tamper-evident event record in a controlled POC.

Evidence boundary Identity uses a development provider; evidence-sealing primitives are verified, but the live verdict pipeline does not yet invoke the sealer.

GenAI Prompt DLP

Available for POC
Module
Sentry
Availability
Available for POC

Evaluate deterministic detection and governed verdicts before an approved client submits text to an AI destination.

Evidence boundary Endpoint interception is simulated and model classification is disabled. The browser demo is not evidence of fleet capture.

AI Usage Observability

Coming next
Module
Cartographer
Availability
Coming next

The engineering build validates normalisation and projection behaviour using synthetic scheduled-file and CEF/syslog-shaped observations.

Evidence boundary Not in the current POC offer. Live source integration, coverage, and completeness are not proven.

Governed decision path

The model does not become the authority

A POC decision follows a legible sequence: simulated input, deterministic detection, policy resolution, governed outcome, and an event record.

Customer-controlled evaluation boundary
  1. Simulated prompt input
  2. Deterministic detector
  3. Versioned policy resolution
  4. Prohibited path terminates
  5. Reviewable event record
Warning issued Available for POC
Authority
Versioned policy resolution
Provenance
Simulated prompt input
Event time
Policy reference
POC-POLICY-EXAMPLE
Simulated input Integrity not asserted for this illustration

POC architecture

Implemented controls, with their edges exposed

The overview names what current code supports and where production integrations remain acceptance criteria.

Identity and roles

Role enforcement is implemented at POC depth through a disclosed development identity provider. Real AD, Kerberos, SPNEGO, PIV, and FIDO2 are absent.

Deterministic policy

The same policy engine drives enforcement, dry run, and effective-policy explanation. Immutable versions make the applied rule set inspectable.

Tamper-evident record

Stored canonical events form a hash chain with signed checkpoints and can be independently verified. SIEM acknowledgement remains emulated.

Evidence custody

Evidence-sealing and custody primitives are test-vector verified. The live verdict-to-sealing pipeline is incomplete and POC custody is software-based.

Resolution trace
  1. 1 Exception
  2. 2 Rollout stage
  3. 3 Rule action
  4. 4 Resolved outcome
Evidence plate

One policy engine, one explainable result

Verified in current build

The same deterministic policy engine drives enforcement, dry run, and effective-policy explanation in the tested POC code.

This verifies policy consistency, not endpoint capture, production identity, model quality, rated latency, or customer deployment readiness.

Material limitations

A POC should test the gaps, not conceal them

These limitations define the evaluation boundary.

  1. 01 No endpoint agent or browser extension exists; capture is simulated.
  2. 02 The model classifier is disabled and no model-quality or language-coverage claim is permitted.
  3. 03 Real AD/Kerberos authentication and hardware-backed evidence custody are absent.
  4. 04 Rated-QPS latency, production capacity, HA, and failover have not been measured.
  5. 05 Runtime zero-egress is a design and acceptance objective, not a proven appliance claim.

Fit by operating need

Three problems the current platform can help evaluate

These are paths through the same current POC, not separate products or claims of production coverage.

Control public-AI data exposure

Test deterministic detection and policy outcomes against approved synthetic or reviewed text samples.

Current POC: Core + Sentry. Endpoint interception remains a disclosed stand-in.

Apply controls inside your boundary

Examine a self-hosted design where policy remains authoritative and model-shaped work cannot silently replace a governed decision.

Air-gap-capable design; runtime zero-egress still requires evaluation proof.

Produce reviewable policy evidence

Inspect versioned policy outcomes and tamper-evident event-chain behaviour while keeping pipeline limitations explicit.

Policy and audit-chain packages are verified only at their stated scope.

Portfolio horizon

Future specialists stay dormant until validated

The remaining modules are design-partner roadmap directions. They do not emit live verdicts, receipts, metrics, or customer outcomes.

  • Email & Channel DLP — Postmaster Design-partner roadmap
  • Private SOC Copilot — Analyst Design-partner roadmap
  • Security Log Intelligence — Correlator Design-partner roadmap
  • AI Application Security — Reviewer Design-partner roadmap
  • Data & Identity Security — Steward Design-partner roadmap
  • AI Governance & Compliance — Scribe Design-partner roadmap
  • Threat, Phishing & Malware Intelligence — Forager Design-partner roadmap

Evaluate the boundary before buying the promise

A private evaluation starts with architecture, data-boundary, and test-case review. It does not ask you to upload sensitive material through this public portal.

Request a Private Evaluation

Further reading Verdict path Maturity ladder Platform Core note