Skip to content

Regulatory compliance

Local-only inference materially reduces the security-review surface for AI-assisted execution in regulated environments.

Because Flow’s inference is local:

  • No cloud data processing agreements (DPAs). Inference data never crosses the workstation boundary, so there is no third-party processor to contract with.
  • No external data-flow audits. There is no external data flow to audit on the inference path.
  • No third-party inference endpoint certification. No external endpoint is involved.

The result is a significantly reduced security-review scope for AI-assisted execution, along with compatibility with on-prem and air-gapped deployments.

Standard organizational IT sign-off for desktop software installation is still required, especially in regulated environments. Workstation-level controls continue to apply. These include endpoint protection, OS hardening, and credential-manager configuration.

Even with zero egress, the following remain in scope for compliance review:

  • Credential custody. The host process is the sole custodian, and it uses OS-native credential stores. See Credentials and PII.
  • Privilege segmentation. See Isolation boundaries.
  • Execution history. Flow runs and node outcomes are stored locally. Cloud-AI nodes persist metadata-only output by default.
  • PII handling. The sanitizer redacts patterns before any text reaches a model.

When Flow scales to enterprise deployment, the zero-egress boundary is preserved by deliberate design choice:

  • The Governance Service processes flow-graph metadata only. It never sees inference data or spool content.
  • Audit forwarding to a SIEM transmits execution metadata only, such as model versions, pass/fail, and who-ran-what. It never transmits inference inputs or outputs.