Security without compromise. AI without the risk.
Giving AI access to your core ERP shouldn't mean gambling with production data. DataTether enforces a zero-trust execution boundary: zero secrets in prompt memory, strict role-based isolation, and mandatory human sign-off on every mutation.
Enforced in hardware and edge proxy architecture · Not system prompt instructions
POST /sap/opu/odata/sap/API_SALES_ORDER_SRV/A_SalesOrder('4500008123')
Four laws of enterprise AI safety.
These aren't soft guidelines or prompt instructions. They are hard architectural boundaries enforced at the proxy layer before any packet reaches your ERP.
No AI mutates production without human sign-off.
Mandatory human approval gate
The AI can analyze, query, and propose transactions, but it cannot commit them alone. Every write, update, or delete pauses for an authorized human to inspect and sign.
Passwords never touch the AI’s memory.
Hardware-encrypted vault isolation
ERP credentials live exclusively in hardware-encrypted vaults. No prompt, log, client app, or model runtime ever sees or stores your secrets.
The AI has no more access than the human operating it.
Strict ERP role inheritance
An assistant operating for a sales rep can only access records permitted by that rep’s specific SAP PFCG role. Zero backdoor superuser privileges.
Every action leaves an immutable paper trail.
Cryptographic audit accountability
Every tool invocation, human decision, and atomic execution is cryptographically signed and streamed to your enterprise SIEM in real time.
What happens between an AI prompt and your database.
Every AI request is intercepted by DataTether before your ERP receives it. Identity is verified, operation scope is checked, secrets are resolved, and mutations pause for human authorization.
Authenticate
Validate caller identity, organization membership, and cryptographic token signature via OAuth 2.1.
Authorize
Enforce tenant isolation boundaries and verify underlying ERP role allowances for the requesting user.
Resolve
Fetch connection secrets on-demand from encrypted vault enclave inside protected Edge Functions with zero memory caching.
Classify
Tag requested operation as safe deterministic read, gated atomic mutation, or multi-step workflow.
Gate & Review
Pause mutating calls; present visual confirmation diff card to authorized human approver before execution.
Atomic Audit
Execute atomic $batch changeset with automatic rollback and stream cryptographically signed audit log to SIEM.
The Proxy Invariant: AI models never obtain direct ODBC, RFC, or unrestricted database credentials. By routing all interactions through DataTether’s proxy edge, every interaction produces an auditable SHA-256 HMAC event before execution.
Zero passwords in prompt memory. Period.
Zero ERP secrets in prompt context or client devices. Full identity federation with your existing corporate Okta or Entra ID tenant.
Credential Storage & Vault Pattern
AES-256-GCM Hardware EncryptionSecrets are stored in hardware-encrypted Supabase Vault tables. DataTether instances only retain opaque vault reference IDs. Credentials resolve solely inside protected Edge Functions at request time and are never written to disk or logs.
Authentication & Identity Modes
Okta, Entra ID, & OAuth 2.1DataTether integrates directly into enterprise identity architectures to ensure every AI invocation carries an authenticated, auditable principal.
OAuth 2.1 Bearer Token
RFC 6749 · 8414Standards-compliant enterprise MCP authentication with scoped role claims and cryptographically signed JWTs.
Corporate SSO & JWT Mapping
SAML · OIDCReuses your existing enterprise identity provider (Okta, Microsoft Entra ID) to enforce individual user authorizations.
Basic Auth with Vault Isolation
Hardware EnclaveSafely maps legacy NetWeaver credentials through hardware vault resolution without runtime or log exposure.
Scoped Machine Tokens
mTLS PinningCryptographically bounded machine tokens with granular entity allowances, rate limiting, and instant kill-switch revocation.
Two ways to connect AI to ERP. Only one survives an audit.
The difference between an enterprise-grade AI architecture that passes SOC 2 and SOX compliance audits, versus a dangerous unmonitored backdoor into your ERP.
What your CISO is actually thinking.
Technical answers for infosec teams evaluating zero-trust Model Context Protocol deployment on SAP.
Schedule Technical Security ReviewDon’t wait for an audit to fix AI security. Test in fifteen minutes.
Hardware-encrypted isolation. Strict role inheritance. Zero blind writes.
We will help you turn a real read or approval workflow into governed AI actions, reviewable UI, reusable prompts, and confirmation gates before any sensitive write can run.
A good first step is small: one service, one user outcome, one visible control boundary.
Months of dev time eliminated
Auto-generated tools replace weeks of custom MCP server development.
No infrastructure to run
Fully managed. Nothing to install, host, or maintain.
Instant dashboards and UI
Live tables, KPIs, and review screens — no frontend developer needed.
One setup, every AI client
ChatGPT, Claude, Manus, custom agents — all from the same connection.
Security is not a feature. It is the platform.
Auto-Discovery & Capabilities
See how raw OData EDMX translates into governed MCP tools and typed schemas in minutes.
02 Interactive UIMCP UI In-Chat Views
Render live sortable tables and approval cards while shielding LLM context tokens.
03 Business ProcessMulti-Step Workflows
Chain ERP reads, policy branches, human approvals, and atomic writes into a single tool.