Review-Ready for
IT and Security
CRAFT is built to pass your security review, not route around it. Local execution, no cloud processing of project data, cryptographically signed artifacts, a fail-closed Gate, and a published vulnerability disclosure policy. Here’s what you need to approve deployment.
Sound Familiar?
"Finance wants us to evaluate this CAD automation tool. I need to know what it sends to the cloud, what it touches on the workstation, how we patch it, how we revoke access, and whether there’s a security contact I can escalate to. I don’t have time to decompile a .NET add-in to find out."
Deployment Summary
The short version of what to expect on the network, the endpoint, and at runtime.
Local Execution Only
Automations run inside the Revit or Civil 3D process on the user’s workstation. There is no cloud execution path, no remote code evaluation, and no server-side processing of project files, model data, or drawing content.
Minimal Network Surface
Outbound HTTPS only, to api.cadcraft.ai
for license activation, token refresh (max once per 7 days), and
signed artifact download. No inbound connections, no daemons, no
background services. See IT / Admin Notes for
the full list.
No Elevated Privileges
CRAFT installs as a standard Autodesk host add-in. No system services, no kernel drivers, no scheduled tasks, no elevated runtime. Installation follows the same privilege model as any other Revit or Civil 3D add-in.
Signed Artifacts, Fail-Closed Gate
Every automation artifact is SHA-256 content-addressed and ES256 signed. The local client verifies signatures against published trust anchors before any import or execution. The Preview/Validate Gate is fail-closed — if it cannot prove an artifact is safe, it blocks execution. No user override.
No Telemetry by Default
CRAFT does not collect or send telemetry unless the user explicitly opts in. Crash reports, usage metrics, and analytics are all disabled by default. Egress disclosure on every operation shows what (if anything) is leaving the machine before it leaves.
Offline for 7 Days
After activation, CRAFT operates fully offline for up to 7 days between license token refreshes. Locally cached artifacts run without connectivity. Air-gapped environments can receive artifacts via exported zip packages and verify them locally.
Security Review Checklist
What a typical IT and security review looks for, with where to find each answer.
Data flow and egress
Every network call is listed in the IT Admin Notes and enforced by the egress disclosure panel in the Gate. Project data does not leave the workstation by default.
Installation and patching
See Installation and Activation. Updates follow semantic versioning; once published, artifact versions are immutable.
Cryptographic integrity
Artifacts are ES256 signed and SHA-256 content-addressed. Trust anchors are available via a public, unauthenticated API for independent verification. See Artifact Integrity.
Tenant isolation and access control
Tenant-scoped authentication and strict isolation server-side. Seat-based entitlements (Runner, Creator, Approver). See Licensing and Authentication & Isolation.
Data protection and privacy
See Privacy Policy. Subprocessors are listed, international transfer mechanisms are documented (SCCs, DPF), and a DPA is available on request at privacy@cadcraft.ai.
Vulnerability disclosure
See Report a Vulnerability. Scoped, with a safe harbor for good-faith research. Machine-readable contact at /.well-known/security.txt. Reports to security@cadcraft.ai.
Audit trails
Every published automation has a full audit history: author, publish time, version, signature, and Gate validation record. Immutable and cryptographically verifiable.
Revocation and offboarding
Seats can be revoked from the licensing service; the Gate hard-blocks seat-required operations on the next refresh (within 7 days). For immediate revocation, contact support@cadcraft.ai.
Where AI Is and Isn't
The boundary, stated plainly, so your AI use policy can map to it.
CRAFT uses AI to interpret task descriptions, search and rank library reuse candidates, and draft test candidates and capability contracts for human approval. AI never writes the executable plan, never modifies script contents, never bypasses the Gate, and is never the sole publisher.
If your organization has an AI use policy that restricts where proprietary data can be sent for LLM processing, the Preview/Validate Gate's egress disclosure panel shows exactly what data (if any) a given automation sends to which destination before it runs. See Security & Trust ยง AI Scope for the full enforcement model.
Deep Dive References
Where to go when your reviewer wants the technical detail.
IT / Admin Notes
Deployment, network behavior, offline operation, artifact integrity, licensing architecture.
Read the notesSecurity & Trust
Full trust model, Gate details, egress disclosure, vulnerability disclosure policy.
Full detailsPrivacy & DPA
Data processing, subprocessors, international transfers, Data Processing Addendum on request.
Privacy PolicyNeed a Walkthrough for Your Review?
We'll walk your security and IT teams through the trust model, deployment architecture, and data-handling posture on a call. Early access participants get this as part of onboarding.