Veritas – Execution integrity runtime for AI agents

Designed Veritas as a runtime that compiles agent conversations into deterministic execution graphs, persists versioned run history, and records explicit decision resolutions. Implemented policy gating and replay verification so the same inputs and decision resolutions produce the same execution graph, surfacing mismatches as integrity errors. Built a browser-based control plane using React and TypeScript with dashboards, run inspection, audit search, approvals, exports, and operations. Implemented multi-tenant isolation with organisation-scoped runs, decisions, and audit logs, and an API key system for creating, rotating, and revoking keys. Added durable background workers for webhooks and export jobs backed by database leases to support concurrent workers and safe recovery.