Trust
Governance you can point at, not adjectives about enterprise AI.
Every control below is a product behavior. Where a capability depends on deployment, this site says so.
Read-only by default. Source-grounded. Human-approved where actions matter.
Controls
Four boundaries that hold in every workflow
Approved sources only
Agents read the schemas, tables, and columns a deployment approves — not whatever they can reach.
Tenant-scoped permissions
Data, dashboard sessions, artifacts, and saved knowledge stay inside the organization and the user's role.
Governed read-only queries
Read-only credentials, one bounded SELECT, blocked mutations, statement timeouts, bounded row counts.
Human approval before it matters
Messages stay unsent, invoices unposted, purchase orders unissued, inventory unchanged, machine controls untouched.
Read-only by default
Operational database tools run under constraints the application enforces, not constraints the model is asked to respect.
- Read-only database credentials
- Allowed schemas, tables, and columns only
- One bounded SELECT or WITH ... SELECT per query
- Mutation blocking
- Statement timeouts
- Bounded row counts
Tenant and identity boundaries
One company's plant data is never pooled into another company's operating context.
- Organizations with users, roles, and invitations
- Tenant-scoped operational data
- Tenant-scoped dashboard sessions and artifacts
- Owner-scoped saved knowledge and views
- Delegated user access for connected systems
- Audit attribution for agent and user activity
Human approval where actions matter
The platform can prepare invoices, purchase orders, reminders, customer updates, operations messages, allocation plans, and maintenance actions. The current boundary is review-first.
- Messages remain unsent
- Invoices remain unposted
- Purchase orders remain unissued
- Inventory remains unchanged
- Payments remain unposted
- Machine controls remain untouched
Delegated connected-system access
For Microsoft 365 and similar systems, the target model is delegated access under the signed-in operator's own permissions.
- The operator signs in to the connected system
- The integration validates operator identity
- The connected system returns only resources that operator may access
- Tokens are not accepted in prompts or tool arguments
- Raw credentials are not exposed in results
- Activity is auditable
Evidence
What can be preserved with a result
- Source system
- Source table
- Source record ID
- Tenant
- Source or asset key
- Event timestamp
- Returned rows
- Generated query
- Calculation steps
- Artifact ID
- Model and agent activity
- Freshness status
Failure behavior
What the product will not do to look better
- Rows are never fabricated.
- An explicit time window is never silently broadened.
- Demonstration data never stands in for an empty live result.
- Planning and query failures are surfaced, not hidden behind an unrelated fallback.
- Wrong output is not patched with prompt-specific title or string rewriting.
- Latest-available historical data is never presented as current.
Disclosure
What we are not claiming
- No security or compliance certification is claimed on this site. If a certification is achieved, it will be listed here with its scope and date.
- No customer names, logos, testimonials, deployment counts, or performance results are published. Where proof would normally sit, this site leaves it out rather than approximating it.
- Interface visuals on this site are labeled composites. They show structure, not plant values.
- Connector coverage, Microsoft 365 access, ERP adapters, and live market feeds are established and validated per deployment — not assumed on day one.
- Autonomous plant control, automatic product-mix or setpoint changes, and automatic posting to business systems are not offered.
Bring your security and IT team to the first call.
Read-only credentials, tenant scope, delegated access, and audit attribution are easier to evaluate against your requirements than against a brochure. Questions before then: team@importintel.net.