PDAccess adds Automated Compliance — PCI-DSS, SOC 2, ISO 27001, HIPAA, NIST, and DORA enforced as deterministic policy at the proxy, with your own policies, RAG-powered runbooks, and an MCP server so AI agents comply too.
Every PAM platform can tell you what happened — a recorded session, a named user, a timestamp. What they can't tell you is whether what happened was allowed. That's the gap auditors probe: "show me that nobody had standing access to production," "show me that every privileged action was attributed to one identity," "show me access was time-boxed."
Today, that answer lives in spreadsheets, screenshots, and hope. Shared accounts, standing privileges, policies documented in a wiki and enforced by nobody.
PDAccess closes the gap by moving compliance from the reporting layer into the enforcement layer. Automated Compliance is a policy engine that applies your frameworks' exact constraints — deterministically, at the tunnel and proxy — before a privileged session is allowed to exist.
The platform ships with a managed reference corpus covering the frameworks your auditors ask for:
Frameworks bind to tags and subnets (tag:env=prod-cde, 10.240.0.0/16), and when a target falls under several, the engine applies the strictest constraint — shortest window, shortest idle timeout, mandatory MFA, forced read-only. No interpretation, no gaps.
Frameworks are the floor. The rule builder lets your compliance admins author their own deterministic policies — max_duration_minutes, idle_timeout_minutes, ticket-ID requirements with regex patterns, default_read_only (mandatory for AI agents), allowed_protocols, MFA, and approval modes from auto_approve to dual_custody. Every rule is JSONB, attached to scope, and enforced at the tunnel.
The unstructured half of compliance — "how do we actually do a production DB migration?" — is handled by RAG. Your SOPs and runbooks (Markdown/PDF) are chunked, embedded, and indexed under your tenant by a CPU-only ONNX pipeline: no external API calls, no data leaving your deployment. Policy answers come back cited to the exact clause or runbook section — the same answers your humans get, your AI agents can get too.
The pdaccess MCP server exposes the whole compliance surface to local coding agents: list_accessible_targets, lookup_compliance_policy, request_jit_access, get_tunnel_status, revoke_jit_access, and generate_compliance_report. An agent asks what it may do, gets the active constraints with citations, requests a JIT lease, and — on approval — the local daemon injects the scoped route and the Access Proxy brokers the flow. Read-only by default, every action attributed to a named agent identity, every request in the same audit trail as human access.
generate_compliance_report aggregates the deterministic proxy audit logs — MFA verification, human vs. agent traffic, full SQL/SSH transcripts — and maps them to the exact clauses of your active frameworks. Multi-tenant isolation (row-level security + tenant-scoped vector filtering) means your policies, runbooks, and audit data never cross tenant boundaries.
Compliance stops being a quarterly reconstruction and becomes the behavior you already have. That's the whole point.