Security overview
AI Solutions Exchange uses multiple layers of control — authentication, role-based authorization, organization membership, database grants, row-level security, server-only privileged operations, API-key verification, protected solution storage, buyer-safe delivery policies, and audit logging. No single control is described as making unauthorized access impossible.
Layered access control
Authentication, role checks (via a private schema helper), organization membership, RLS, and server-only privileged operations — not any single control on its own.
Protected solution storage
Proprietary solution content lives in a deny-by-default table with forced RLS. No buyer-facing policies. All writes go through authorized server functions.
Buyer-safe delivery
For protected listings, setup submission returns readiness, warnings, and permitted next actions — not the underlying prompt template.
Unified solution manifest
One schema powers the marketplace, approval packets, public API, and future editor integrations. Buyer-safe fields only in public responses.
Server-controlled promotion
A solution becomes protected only after server-side readiness evaluation. Selecting a UI option does not grant protected status.
Additive schema, audited actions
Migrations are additive. Privileged actions are recorded in audit_logs with actor, action, target, and timestamp.
Current implementation status
A transparent view of what is confirmed in the current codebase, what is partial, and what is planned. We do not claim SOC 2, ISO 27001, penetration testing, or regulatory approval — see the FAQ for how we talk about compliance today.
Implemented
10- •Unified solution manifest schema
- •Protected content table with forced RLS and no buyer policies
- •Authorized server functions for protected content writes
- •Buyer-safe wizard delivery policy for protected listings
- •Public manifest with buyer-safe posture fields
- •AI analysis, certification, and setup-wizard generation
- •API keys (SHA-256 hashed, shown once)
- •Organizations, memberships, and policy records
- •Audit log for privileged actions
- •Approval packets and enterprise documentation generator
Partial
4- •Protected execution runtime (storage & delivery in place; full session runtime under development)
- •License entitlement expansion
- •Organization policy evaluation across all deployment paths
- •Data-retention configuration surface
Planned
6- •Session authorization, nonces, and streaming for protected AI sessions
- •Published editor extensions (VS Code, Cursor, Windsurf, JetBrains)
- •Organization-controlled AI gateways
- •Stripe Connect payouts
- •Independent penetration testing when commissioned
- •Formal compliance programs (e.g. SOC 2)
Protected Solution Architecture
Buyers primarily license the ability to configure and use an AI solution. For protected listings, the underlying proprietary content is not returned through the marketplace UI, setup flow, public manifest, approval packet, or buyer-facing API response. This reduces unnecessary exposure of proprietary logic and makes unauthorized redistribution more difficult than delivering raw prompt content — it does not guarantee that behavior cannot be inferred from outputs.
Organization-governed marketplace access
When an organization uses AI Solutions Exchange, its administrators define the models, capabilities, data-handling requirements, deployment modes, and risk levels permitted within the organization. Each solution's unified manifest is evaluated against those rules. Only solutions that satisfy the organization's policy — or receive the required internal approval — are available for purchase and deployment by its members, and the planned IDE extensions will render only that organization-approved catalog for ordinary members.
Descriptive tags support classification. The enforceable gate is the comparison between structured manifest fields and the organization's policy — not tags alone.
Shared responsibility
Platform, solution architects, buyers, organizations, and model providers each own part of the security picture. See the FAQ for the full breakdown.
Approval Packet
Every listing can generate an executive-ready packet consolidating manifest, permissions, risk, dependencies, compatibility, setup estimate, and certification results.