NovaHR is an AI "HR digital employee" demo, and the useful part is not the chatbot — it is the workflow. Six HR workspaces sit behind a single shell, and each one plays out a real process end to end: screening, onboarding, training, review, policy questions.

The front end is React and Vite; a small Hono adapter in TypeScript sits between it and the data and runs in two modes. Mock mode is the default and needs no keys: every action replays a recorded fixture and the response declares which mode produced it. Real mode spawns actual agent runs through an OpenClaw profile and needs the CLI plus a profile directory holding credentials, neither of which is bundled — if the profile is missing the launcher warns and falls back to the fixtures.

The persona is Nova, a digital HR colleague. Recruiting parses and scores a batch of resumes, searches the talent pool by semantic similarity and triggers interview minutes. Onboarding runs five parallel tracks behind an approval gate. Training turns one topic into a needs analysis, slide outline, instructor manual, student handbook, quiz and evaluation form. Performance aggregates 360-degree feedback into a competency radar and flags blind spots. Employee relations answers policy questions with clause-level citations and escalates red lines. An orchestration console dispatches natural language to the right workspace under three permission tiers, and an asset library collects every document the run produced.

Read it for the process modelling. Each workspace is a flow with state, approvals and generated artefacts, played back step by step over server-sent events — so you can watch which step produced which document, and where a human was allowed to step in.