Security
What this site stores, how the platform treats commands and custody records, and how to reach us about a vulnerability. Plain language, no theater.
What the engine enforces
These are defaults in the coordination runtime, not policies. The public demo reports them live under Runtime internals on /demo.
Evidence: 25 tests in tests/test_authority_transfer.py
(pytest --collect-only -q reports tests/test_authority_transfer.py: 25),
and 209 checks in tests/test_integration.py, which that file runs itself
(pytest --collect-only -q reports tests/test_integration.py: 1,
because those checks are not named test_*). Sixty-six further gate tests — e-stop during handoff, durable store, actuation choke,
power-loss, physical-control countersign, and receipts — run in the same suite. A written threat model and
evidence register sit in the repository. All of it runs against a simulated fleet. No physical
robot has been operated through this system, and this page does not claim otherwise.
No accounts. The waitlist is the only PII, and it is documented
swarmonic.ai has no login and no user accounts, and it keeps no password database. The public demo does not require an account. The waitlist collects email, name, organization, role, and whether you own robots — that is PII, and it is documented on /privacy. A Vermarco purchase receipt records the sale. After its signature verifies, fulfillment issues a sim-fleet operator token and stores only the token's hash. The receipt and the token are not robot-control credentials. No page on this site commands physical hardware. Interactive API explorers are off. Public routes are read-only fleet and portfolio metadata, plus named POSTs for the waitlist, onboarding, license fulfillment, and the demo's emergency-stop, resume, and coordination scenarios.
Encryption in transit
The app redirects a www host to the https apex. TLS for swarmonic.ai is
terminated at the edge, not in this process. The service worker caches
same-origin pages and static assets for offline fallback and does not cache
/api or /ws. No telemetry or analytics beacons run
on this site.
Response headers you can verify with one request
Content-Security-Policy— self-origin script/style/connect policy.X-Frame-Options: DENY— no framing.Referrer-Policy: strict-origin-when-cross-originX-Content-Type-Options: nosniffPermissions-Policy— camera, microphone, and geolocation disabled.
Signed, attributable, tamper-evident
The orchestration platform is designed so that every command is attributable: signed command records, named refusals, override and suppression records, and hash-linked custody logs. Emergency stop is never gated by any consent flow, permission tier, or platform state. These properties are enforced by the automated tests in the repository.
What emergency stop is, and is not. emergency_stop in Swarmonic is a supervisory command sent over the control protocol. It is not a safety-rated function under ISO 13849 or IEC 61508, and Swarmonic declares no Performance Level (PL) or Safety Integrity Level (SIL). The safety-rated stop for every robot remains the manufacturer's certified hardware stop circuit and the owner's on-site protective devices, which Swarmonic never bypasses, gates, or replaces.
Report a vulnerability
If you find a security issue on this site or in the platform, email hello@swarmonic.ai with steps to reproduce. We acknowledge good-faith reports, we do not pursue researchers acting in good faith, and we credit fixes on request. Please do not test against systems you do not own.
Deployments are governed by signed agreements
Production deployments run under signed agreements with their own security schedules, which supersede this page where they conflict. This page describes the public site and the platform's design posture. Last updated August 15, 2026.