Enterprise Guardrails
Trust is paramount when giving AI access to your data. Stellar Intelligence X is architected to ensure that your database is completely isolated and conversational queries remain strictly read-only.
The Secure Request Lifecycle
Schema Extraction
Stellar retrieves only schema metadata (table names, column names, semantic descriptions). Raw row data is never transmitted to the AI engine.
Guardrail Verification
Before execution, our strictly defined Mathematical AST Guardrail layer intercepts the query. Hallucinations or malicious injections are neutralized instantly.
Strict Read-Only Enforcement
The Guardrail layer scans incoming SQL for destructive keywords. Any operation outside of the safe `SELECT` scope is rejected immediately.
Execution Architectures
We enforce isolation strategies depending on your deployment choice: Federated or Direct.
Direct Database Binding
We utilize sqlglot to parse AI-generated SQL into an Abstract Syntax Tree (AST), programmatically injecting your Nested Governance rules into the outermost WHERE nodes safely.
WHERE (category = 'Electronics')
AND (tenant_id = 'org_123') -- AST Injected Security Node
Zero-Credential Federated Hooks
If enabled, Stellar doesn't execute SQL at all. We transmit the abstracted AST logical groupings securely (via HMAC-SHA256 post hooks) to your local backend. You maintain 100% control over interpretation and query generation on your side.
Server-Side Capability Gating
Features like AI Insights and Metadata visibility are gated at the server level. Even if a user attempts to enable them via client-side attributes, the backend will strictly suppress restricted data based on your Workspace Whitelabel configuration.
Isolate at the Database Layer
Always connect Stellar Intelligence X using a database user that has been explicitly granted strictly READ ONLY permissions in your backend system. Even with guardrails, this is non-negotiable for enterprise deployments.