ComplianceLint

ComplianceLint runs EU AI Act compliance scans directly inside supported IDEs — Cursor, Windsurf, Claude Code, Zed, Copilot, Codex, and any MCP-compatible environment. Setup is a single command (`npx compliancelint init`) or a Python CLI alternative. From there, a developer asks their AI assistant to scan the project; the tool reads files locally, maps results against 247 legal obligations across 44 articles, and returns a finding breakdown — compliant, non-compliant, and needs-review — in seconds. Remediation tasks are generated in the same session, prioritized by article, and tracked across re-scans so the team can watch their compliance score move as fixes land.

The differentiating constraint is local execution. The vendor states explicitly that code never leaves the machine, which matters for teams building in regulated sectors — AI finance, AI medical — where uploading source to a third-party SaaS for compliance review creates its own legal exposure. Findings sync to a cloud dashboard only as structured metadata, not raw code. The audit trail records who attested each finding and when, which is the artifact auditors under the EU AI Act will actually ask for.

The tool fits AI product teams that need to start compliance work before they have a legal team engaged, and organizations preparing for a formal audit who need something other than a manually maintained checklist. Where it breaks: the ‘needs review’ category is not a machine-resolvable state. Human oversight obligations, role-specific attestations, and Article 14 requirements around human control require the team to work through guided forms — the scan surfaces the obligation, but the compliance work is still a human decision. Teams with complex multi-repo architectures will also find the dashboard’s cross-repo scoring useful but will need to manage attestation consistency across repositories manually.

License is BSL 1.1 (source-available, not open-source), and the tool is scoped to B2B use — the vendor’s terms exclude individual consumer use. Self-hosted operation is supported via the local scan model. A paid-only tier exists beyond the free starting point, though feature gating between tiers is not detailed on the public page.