What ORP is

ORP is a device-first messaging protocol whose rendezvous server is structurally unable to read message contents, keys, or unsealed signaling. The server brokers a meeting between two devices that already share one key bound to one target, and that blindness is an enforced invariant of the wire format, not a logging policy the operator promises to follow.

The one-key-one-target model

ORP binds a shared key to a single rendezvous target. Possessing the key lets exactly two parties meet at exactly one point, and nothing wider. There is no broadcast surface and no directory the board can mine: a key opens one door and only that door.

The blindness invariant in plain terms

The board (the rendezvous server) never receives readable contents, private keys, or unsealed signaling. What crosses it is reduced to opaque presence and routing tags plus sealed payloads it cannot open. Because no frame shape carries plaintext the board could read, blindness does not depend on the operator choosing not to log. Remove the operator’s good behavior from the model entirely and the property still holds.

What the board does and does not see

• Does see: opaque presence tags, a routing tag (frame_kind) that lets it forward frames, and sealed payloads it cannot open. It can therefore observe that two endpoints rendezvous (the social graph, see the threat model).
• Does not see: message contents, private keys, or unsealed WebRTC signaling (SDP and ICE).

Where to go next

• Quickstart: install, test, and run the demo.
• White paper: the full protocol design, threat model, and verification methodology in one document.
• Protocol reference: wire types and the KEY/offer/answer phases.
• The blindness invariant: the collision analysis behind one key, one target.
• Threat model: adversaries, protections, and the stated trade-offs.

Source: SPEC §0/§1 (blindness invariant, one-key-one-target), SPEC §7 (single stateful channel). TODO: confirm section numbers against the upstream spec (see OPEN-QUESTIONS.md).