Record by ~2:30pm as fallback if live demo fails on stage.
- Terminal:
POLICYGUARD_BASE_URL=http://localhost:3000 npm run demo(or deployed URL) - Optional: ClickHouse / Senso GEO tab
- Browser: https://policyguard-site.vercel.app
AI agents are taking real actions on the open web. Computer Use, Operator, custom scrapers — they hit "am I allowed to do this?" hundreds of times a day. Today nobody checks. Agents get banned. Lawsuits happen.
PolicyGuard is the public compliance layer for the agentic web. Watch.
Run npm run demo or call API three times. Narrate each line:
| # | Say | Expect on screen |
|---|---|---|
| 1 | "Sales agent wants to scrape 100 LinkedIn profiles." | decision: blocked, no_bots, citation from LinkedIn ToS |
| 2 | "Procurement bot reads OpenAI public pricing once." | decision: allowed, low risk |
| 3 | "Same agent tries to bulk-store emails in HubSpot." | decision: modify_recommended, requires_human_review: true, safe alternative |
Point out: matched_rules (machine-readable), machine_instruction (executable flags), cited_md_url (published corpus).
Every decision is logged and published to cited.md. The corpus compounds — the next agent can cite our prior answer. We're building shared infrastructure for the agentic web, not a policy summarizer.
- Full terminal output visible for all 3 scenarios
- Verdict JSON shows three different
decisionvalues - Mention x402 payment if Aslan has it wired (optional clip)
- Show https://geo.senso.ai KB briefly (Senso prize)
- Save video file named
policyguard-demo-backup-2026-05-23.mp4
npm run dev
# other terminal:
POLICYGUARD_DEMO_MODE=true npm run demo
# live path (needs keys):
POLICYGUARD_DEMO_MODE=false POLICYGUARD_SKIP_PUBLISH=true npm run demo