What we do, what we don't, and how to reach us when something is wrong.

The runtime surface, honestly described.

These are capabilities AegisGrid provides in the deployed runtime. None of them are certifications; they are architectural commitments.

• Local inference.

  All reasoning runs on customer-operated hardware (Ollama or vLLM). No inference request leaves the host.

• Hash-chained, signed audit log.

  Every agent tool call, prompt, and decision is logged, signed, and replayable.

• RBAC + break-glass admin.

  Role-based access control scopes who can invoke workflows; a designated break-glass role holds the containment switch.

• mTLS between agents.

  All agent-to-agent and agent-to-tool traffic is mutually authenticated and encrypted in transit.

• On-prem, air-gapped, or private cloud.

  Three deployment topologies. Air-gapped is the most restrictive; private cloud is the most common.

Explicit non-claims. Every one of these is binding.

We list these because the absence of the claim matters as much as the presence of the claim.

• No SOC 2 Type II audit report yet.

  AegisGrid does not currently hold a SOC 2 Type II audit report. NetWit parent is undergoing SOC 2 Type II audit; AegisGrid will inherit those controls when available.

• No transmission of your data to commercial APIs.

  There is no path by which your documents, alerts, or extracted controls reach OpenAI, Anthropic, Google, or any other commercial inference provider.

• No training on your data.

  We do not train models on customer artifacts. There is no shadow training corpus built from your vendor data.

• No autonomous vendor approval decisions.

  AegisGrid produces a draft risk matrix. The human analyst reviews, edits, and signs off. We never make a vendor go / no-go decision.

• No retained document beyond the assessment record.

  Documents are auto-deleted 30 days after the active assessment unless you request an extension. Configurable per engagement.

Where your data lives, who can see it, and for how long.

• TLS 1.3 in transit · AES-256 at rest · mTLS between agents.

• On-prem or private-cloud deployment by default. Air-gapped topology available.

• For Restricted data (e.g., Rutgers Policy 70.1.2 Restricted classification), zero third-party sub-processors; inference is local.

• For non-Restricted data, the only sub-processors are the customer's own ServiceNow instance and SIEM (configurable). Customer is the data controller.

• Retention default: 30 days for uploaded artifacts; 7 years for audit-log entries (configurable per customer compliance regime).