Mindburn Labs
Founding Cohort

Run the Demo

Run HELM's governed company demo — 15 receipts, 7 phases, full EvidencePack in 2 minutes.

SCENARIOGoverned company demomock provider7 phases15 receiptsHTML proof reportPHASE TIMELINE1 Approval2 Execute3 Deny4 Sandbox5 Skill gap6 Incident7 Export + verifyThe demo exercises approval, denial, sandboxing, incident handling, and final EvidencePack verification in one run.
Run the Demo: technical reference for this page.

Run the Demo

See HELM in action with a realistic multi-phase scenario: an AI agent manages a company, hits policy boundaries, and produces a verifiable proof trail.

What you'll see

The demo runs 7 phases that exercise different enforcement properties:

Phase What happens Receipt type
1. Approval Agent requests budget approval → timelock ceremony ALLOW
2. Tool execution Agent calls approved tools ALLOW
3. Policy deny Agent tries a forbidden action DENY
4. Sandbox WASI-sandboxed compute with gas limits ALLOW
5. Skill gap Agent encounters an unregistered tool DENY
6. Incident Agent triggers incident response flow ALLOW
7. Export EvidencePack exported and verified —

Result: 15 receipts forming a complete ProofGraph, plus a shareable HTML proof report.

Run it

# 1. Onboard (if you haven't already)
helm onboard --yes

# 2. Run the company demo with mock provider (no API key needed)
helm demo company --template starter --provider mock

# 3. Export the EvidencePack
helm export --evidence ./data/evidence --out evidence.tar

# 4. Verify offline
helm verify --bundle evidence.tar

Expected output:

verification: PASS

Explore the results

# List all receipts from the demo
helm pack list

# View incident history
helm incident list

# Get a daily brief
helm brief daily

Open the Proof Report

After running the demo, open the generated HTML report:

open data/evidence/run-report.html

The report shows:

  • Causal chain visualization (which receipts link to which)
  • Receipt details (tool name, args hash, verdict, reason code)
  • Verification status per receipt
  • One-click sharing

Run the full use-case suite

For a comprehensive test of all 12 enforcement properties:

make crucible

This runs UC-001 through UC-012:

UC Tests
UC-001 PEP allows valid tool call
UC-002 PEP denies schema mismatch
UC-003 Approval ceremony (timelock + challenge)
UC-004 WASI sandbox executes transform
UC-005 WASI traps on gas/time/memory exhaustion
UC-006 Idempotency (receipt-based dedup)
UC-007 EvidencePack export
UC-008 Replay verify
UC-009 Output schema drift → hard error
UC-010 Trust key rotation + replay
UC-011 Island mode (build offline)
UC-012 Conformance L1 + L2 gates

Next steps