Security & compliance

Security is architected in, not bolted on.

The questions your security team will ask — answered before they ask them. Most of the audit evidence a SOC 2 or ISO 27001 process needs is produced as a byproduct of normal operation, not a special export.

Your keys, your data, your model.

Nothing here is a policy promise — each point below is enforced at the credential, network, or execution layer, not just documented in a handbook.

🔑
No long-lived cloud keys
SnapFlow connects to your cloud via OIDC federation — short-lived, federated credentials assumed per operation. There are no standing access keys to leak.
🎯
Least-privilege, scoped credentials
Workers receive only the scopes a step needs. A task that hasn't been approved cannot obtain elevated credentials — the approval gate is enforced by the credential layer, not just the UI.
🚫
Un-bypassable action gate
Every destructive or permission-changing command is classified server-side and blocked at the point of execution unless a human decision is on record. The model that produced the command cannot skip it.
🧱
Per-tenant isolation
Each agency's infrastructure state and data are isolated — separate state schemas, per-agency scoping on every record.
📦
Isolated execution
Work runs in isolated worker containers on an internal-only network, so the action gate can't be bypassed by direct egress. Read more on how it executes →
🧠
Bring your own model
Your own LLM key is used for your work — no silent fallback to a shared default. Your data goes to the provider you chose. This is possible because SnapFlow's execution core is LLM-agnostic, not built around one vendor.
🤐
Secrets stay secret
Runtime secrets are injected at execution and never exposed in logs, receipts, or the UI.
📜
Full credential audit trail
Every credential issuance and every gated action is recorded in the audit log — who, what, when, and under which approval.
Proof point (live): the connection setup shows the exact OIDC role and scopes granted for your account; the audit log shows every credential issuance and every gated action as it happens — not a quarterly summary.

Built to make an audit a data export, not a fire drill.

Every consequential action becomes structured evidence automatically — you're not scrambling to reconstruct a timeline when an auditor asks for one.

📋
Complete audit trail
Approvals, credential issuance, policy decisions, settlements, and rollbacks are all recorded as structured audit events with actor, target, and timestamp.
🗺️
Control mapping
Security findings are mapped to SOC 2 and ISO 27001 control references, so your evidence lines up with the frameworks auditors already use.
🔏
Tamper-evident receipts
Each mission receipt carries a SHA-256 content hash — outcomes can be independently re-verified, not just asserted.
⚙️
Policy-as-code as continuous control
Guardrails like "production changes require approval" or a weekend change freeze are enforced automatically and logged — the continuous, provable control SOC 2 auditors look for.
Positioning: SnapFlow is compliance-ready by construction — the audit artifacts a SOC 2 / ISO 27001 process needs are produced as a byproduct of normal operation. We state current certification status plainly rather than implying one we don't yet hold.
Control readiness
Controls mapped to the standards enterprise buyers ask for
SOC 2 ISO 27001 CIS Controls v8 GDPR
Self-attested, audit-ready evidence — verified live from our platform. Not a certification.
View Trust Center → security.txt

See the whole loop this protects

Cortex planning, the pre-flight gate, ephemeral containers, and proof-carrying verification — how a mission actually runs.

Read the Platform page →