Compliance Fabric

A continuous-compliance control and evidence layer for regulated Kubernetes platforms. Author controls once, enforce them as policy, and produce audit-ready evidence on demand.

The Fabric targets life sciences first (GxP) and extends to other regulated sectors such as finance (DORA, NIS2). It builds on open standards and open-source projects already common in regulated platforms: OSCAL, Kyverno, OPA, Sigstore, and GitOps.

Status

Active development. The core is a standard-library-only Go CLI (fabric), built test-first, that already runs the continuous-compliance loop end to end:

• Control authoring, policy, and enforcement (Phase 1, done). OSCAL catalogs (EU GMP Annex 11, 21 CFR Part 11), a pharma MES baseline profile, and a native composer (fabric generate) that assembles a Kyverno policy set for the selected controls from the policy library, resolving each control's rules to the policy resources that enforce them and carrying fabric.control-id traceability (it selects and copies reviewed policy resources; it does not author policy bodies from control text). Admission and runtime enforcement are proven on a live kind cluster in CI.
• Trusted delivery (Phase 2, done). SBOM and SLSA provenance evidence producers, Sigstore keyless signature verification at admission, GitOps change-control evidence, and a release evidence gate (fabric release-gate) that binds supply-chain evidence generation into the release pipeline. Proven end to end with syft and cosign in CI. The release manifest declares both an SBOM and a SLSA provenance source, so a tagged release is gated on its own supply-chain evidence and the artifacts, the SBOM, the provenance and the evidence ledger are all signed keylessly with the workflow's OIDC identity.
• Evidence and reporting (Phase 3, done). A hash-chained, append-only evidence ledger; seven evidence producers feeding it; OSCAL assessment-results and a control-posture rollup with a live dashboard; and a continuous collector that polls sources and records only state changes.
• AI agent governance (Phase 4, in progress). A versioned agent/prompt/tool registry, an inline gateway that admits or blocks agent interactions at request time (content guardrails, a model allow-list, input/output screening, and per-agent rate and cost limits), a live LLM/MCP traffic proxy (fabric gateway-proxy) that enforces the same gates on the real model/tool call before it reaches the upstream and screens the model's response on the way back — incrementally, event by event, for streamed (SSE) responses — pre-promotion evaluation gates, and interaction tracing that ingests the gateway log and OpenTelemetry OTLP/JSON. What remains is operational hardening (a reference deployment) rather than new control logic.
• Cross-sector crosswalks (Phase 5, done). A crosswalk engine (fabric crosswalk) that reuses controls the Fabric already enforces to answer a second framework's citations with no new enforcement, with worked DORA, NIS2, and ISO/IEC 42001 crosswalks built from the existing GxP and agent-layer controls. Crosswalks are first-class control-tree artifacts whose referential integrity fabric validate checks.

Phase 4's data-and-logic scope is now complete; what remains is operational hardening (a reference deployment) rather than new control logic. See ROADMAP.md for detail, CHANGELOG.md for recent changes, and CONTRIBUTING.md if you want to help build it.

Open-core

The core is open source under Apache 2.0: the framework, policy templates, OSCAL control mappings, integrations, and documentation. A commercial layer is offered separately by the project's commercial sponsor and is not part of this repository: a hosted control plane, polished audit-pack and validation-report generation, certified GxP control packs, and enterprise support. The boundary is described in GOVERNANCE.md and the reasoning in docs/adr/0004-open-core-and-apache-2-0.md.

The control logic is open on purpose. A quality team and an auditor can read exactly how a control is enforced and evidenced, which is worth more than a black box that asks to be trusted.

The continuous-compliance loop

1. Author controls as code in OSCAL.
2. Compile controls into Kubernetes policy.
3. Enforce at CI, admission, and runtime.
4. Run regulated workloads and AI agents on a qualified platform.
5. Collect platform state and telemetry as evidence.
6. Map evidence back to controls and report. Gaps feed back to step 1.

Documentation

Start at the documentation index, or open a layer directly.

Path	Contents
docs/00-overview.md	Problem, concept, scope, glossary
docs/01-architecture.md	The six layers, data flow, trust boundaries, deployment model
docs/02-control-authoring.md	Controls as code in OSCAL
docs/03-policy-enforcement.md	Compliance-to-Policy, Kyverno, OPA, admission and runtime
docs/04-trusted-delivery.md	SBOM, SLSA, Sigstore, GitOps change control
docs/05-agent-governance.md	AI gateway, agent registry, guardrails, tracing
docs/06-platform-substrate.md	Clusters, workloads, identity, observability
docs/07-evidence-and-audit.md	Continuous assessment, evidence ledger, audit packs
docs/08-regulatory-mapping.md	Control-to-technology mapping
docs/09-validation-approach.md	GAMP 5, CSA, platform versus workload validation
docs/10-build-vs-buy.md	Open source versus proprietary IP
docs/adr/	Architecture decision records
reference/architecture.drawio	The reference architecture diagram

Project files

File	Purpose
LICENSE	Apache License 2.0
NOTICE	Copyright and sponsorship notice
CONTRIBUTING.md	How to contribute, including commit sign-off
CODE_OF_CONDUCT.md	Contributor Covenant 2.1
GOVERNANCE.md	Roles, decisions, and the open-core boundary
MAINTAINERS.md	Who reviews and merges
SECURITY.md	How to report vulnerabilities
ROADMAP.md	Planned phases
CHANGELOG.md	Notable changes

Contributing

Contributions are welcome: control mappings, policy templates, integrations, documentation, and issues. Start with CONTRIBUTING.md and ROADMAP.md, and see GOVERNANCE.md for how decisions are made and how to become a maintainer. All participants follow CODE_OF_CONDUCT.md.

Security

Report vulnerabilities privately as described in SECURITY.md. Do not open public issues for security problems.

What it is not

The Fabric generates and maintains evidence. It does not replace the quality sign-off or the validation decision. That separation matches the risk-based assurance approach in GAMP 5 Second Edition and the FDA Computer Software Assurance (CSA) guidance. See docs/09-validation-approach.md.

License

Apache License 2.0. See LICENSE and NOTICE.