Security & Trust
Security, boundaries and proof by design.
AgenticFy is built to prepare companies for AI agents without taking over payments, orders, inventory, customer data or uncontrolled actions. We are early, and we are explicit about exactly what we read, write, encrypt, log, prove, approve and never touch.
Data handling
What we collect is scoped to preparing your web presence for agents. We do not collect what we do not need.
| Category | What it is | Notes |
|---|
| Domain and public site data | Publicly reachable pages of the domain you assess | Public data; read to score readiness |
| Assessment and proof data | Scores, gaps, evidence references, journey replays | The output of an assessment |
| Generated assets | Entity profile, Q&A, claim registry, manifests | Generated from your assessment |
| Workspace and setup data | Plan, package, owners, setup choices | What drives activation |
| Integration metadata | Provider, connection status, last error | Booleans and status, not secrets |
| Credential secrets | Provider tokens you connect | Encrypted at rest, never returned (see below) |
| Delivery and audit logs | Email/webhook deliveries, admin actions | For operability and audit |
Retention, deletion, export
- ·Retention and deletion are configured by customer agreement; we do not impose a one-size policy here.
- ·You can revoke a connected credential at any time; revocation is supported per provider.
- ·Assessment and proof data can be exported (one-pager, proof package, JSON).
We use third-party APIs and model providers to do this work, so data is processed by those subprocessors (listed below). We do not claim data never leaves the environment.
Credential security
- ·Provider credentials are encrypted at rest (AES-256-GCM) before they are stored.
- ·Secrets are never returned via the API or the UI. Status endpoints return connection state, never the token.
- ·No secrets in logs. We do not write tokens to logs, traces or audit metadata.
- ·Access is scoped by provider, and least privilege is encouraged (read scope is enough for readiness).
- ·Revocation is supported: disconnect a provider and the stored secret is removed.
- ·The admin console never displays tokens; it shows connection status only.
- ·Internal secrets (encryption key, admin secret) live in environment variables, not in the database or the codebase.
| Credential | Used for | Scope we ask for |
|---|
| Shopify admin token | Read products / policies for commerce readiness | Read; write only if you authorize publish |
| VTEX appKey / appToken | Read catalog / SKU / pricing / inventory / policy | Read |
| Webflow token | Publish content when explicitly connected | Content scope only |
| WordPress application password | Create a marked page | Page-create capability only |
| Stripe test secret | Test payment rail only, if configured | Test mode only, never production |
| Google Search Console OAuth | Read search performance signals | Read-only GSC scope |
Permission boundaries
| AgenticFy can | AgenticFy never does |
|---|
| Read your public site | never processes payments |
| Generate agent assets | never moves money |
| Create a GitHub PR (you merge) | never stores card data |
| Publish content when explicitly connected | never creates orders |
| Validate an install | never changes inventory |
| Monitor drift | never changes prices |
| Send signed webhooks | never issues refunds |
| Generate a proof package | never accesses buyer PII unless explicitly scoped in future |
| Surface human approval points | never approves legal/compliance claims without human review |
| never exposes stored tokens |
| never declares readiness without evidence |
Payment boundary
- ✕AgenticFy is not a payment processor and does not store card data.
- ✕Payment, order and inventory operations are blocked by design, not merely discouraged.
Human approval model
- ·Publishing requires an authorized channel and credential; nothing publishes without your connection.
- ·Claims and policies are surfaced for the owner to review; AgenticFy does not self-approve compliance claims.
- ·Payment, order and inventory operations are blocked by design and are explicit non-goals.
- ·High-risk actions either require human approval or are out of scope; the control plane records human approval points.
Evidence and audit
- ·If AgenticFy cannot prove a claim, it marks it as not proven.
- ·Every proof claim carries evidence refs or is marked not proven.
- ·The governance control plane tracks evidence sources and risk.
- ·The proof package shows journeys and failures, not just a score.
- ·Admin actions are written to an audit trail (no secret in the metadata).
- ·Delivery logs record every email/webhook attempt and status.
- ·Webhooks are signed; monitoring history is retained for drift review.
Webhook and integration security
- ·Webhooks are signed with an HMAC over the raw body; verify the signature before trusting a payload.
- ·Delivery uses retry with backoff and a dead-letter state for exhausted attempts.
- ·Deliveries are idempotent: the same event id may arrive more than once.
- ·No credential is ever placed in a webhook payload.
- ·Outbound fetches to customer-supplied URLs are validated to reduce SSRF risk.
Admin and internal operations
- ·Admin routes are protected by an admin secret (a server-side secret, not SSO).
- ·The admin UI does not display secrets or tokens.
- ·Every manual admin action is logged to the audit trail.
- ·Admin cannot force a feature to product-grade or mark a partial as ready.
- ·Plan changes are audited.
Subprocessors
Infrastructure and service providers we use:
- ·Vercel — application hosting.
- ·Supabase — database, authentication and storage (row-level security on; access via scoped server-side functions).
- ·Resend — transactional email, when configured.
- ·Anthropic / OpenAI — language models used to generate and analyze.
- ·Google APIs — Search Console / OAuth, when connected.
- ·Stripe — test payment rail only, if configured; never production payments.
- ·Provider APIs connected by you — Shopify, VTEX, Webflow, WordPress, GitHub.
Specific subprocessors may vary by deployment, customer configuration and connected integrations. We do not claim SOC 2, ISO 27001, PCI or HIPAA certification; we are early and say so plainly.
Security contact
For a security review or to report an issue, contact us and we will respond. A dedicated security contact is available on request.