Blog
Engineering

Proving identity with no authority to ask

Proving identity with no authority to ask

TL;DR

On Offline Protocol, an identity is a keypair generated on the device, so verifying a sender is a local signature check, with or without a network. New keys are trusted on first use, then pinned. Credentials expire, but there is no gossip-carried revocation list today.

Almost every identity system in production assumes you can phone home. A login checks a credential against a server. A certificate is trusted because a chain of authorities vouches for it, and revocation works because clients can reach a list that says which certificates went bad.

Take away the network and all three assumptions fail at once. You cannot reach the server, you cannot walk the chain, and you cannot fetch the revocation list. Offline Protocol operates exactly where those assumptions break, so identity has to work without any of them.

Device-generated keypairs as identity

On Offline Protocol, an identity is a keypair that is generated on the device and never leaves it. The public key is the name; the private key is the proof. The starting point is to stop treating identity as a lookup. There is no registration step with a central authority, because there is no central authority. Two devices that have never seen each other, and have no path to any server, can still establish who they are talking to, because the proof travels with the party making the claim rather than living in a directory somewhere else.

Sessions are built on Ed25519 identities and encrypted with MLS (RFC 9420), the same primitives the transport layer uses end to end. That choice matters here: MLS was designed for group messaging where membership changes and no participant is guaranteed to be reachable at any given moment, which is the offline mesh’s problem restated in a different vocabulary.

Offline signature verification

When device A receives a message that claims to be from device B, it does not ask anyone whether that claim is true. It checks a signature. B signed the message with its private key; A verifies it against B’s public key. The math either holds or it does not, and it holds or fails identically whether A is in a data center or eight BLE hops deep in a building with no coverage. Verification is local, constant-time in the sense that matters operationally, and completely independent of connectivity.

This is what lets identity survive the relay path. A message in the mesh can cross up to eight intermediate devices and switch transports mid-route, and every one of those relays forwards ciphertext it cannot read. The signature the recipient checks is the sender’s, not the last hop’s. Intermediaries move bytes; they do not vouch for anyone, and they are not trusted to.

Trust on first use and key pinning

Offline Protocol uses trust on first use, then pinning, the same answer the SSH ecosystem has relied on for decades. Without an authority to anchor trust, the honest question is: the first time A sees B’s key, why should A believe it belongs to B and not an impostor? A binds the key to the counterpart when they first establish a session, ideally reinforced by an out-of-band factor that physical proximity makes easy, and any later session that presents a different key for the same counterpart is a loud, detectable event rather than a silent substitution.

Proximity is an asset offline systems have that server-based ones do not. Two people standing in the same room, or two vehicles in the same convoy, can confirm a key through a channel an attacker on the far side of the world cannot touch. The mesh turns physical presence into a trust primitive.

Key revocation without a network

For key revocation, Offline Protocol keeps the surface small instead of pretending to solve it: sessions re-key on demand, MLS key packages and issued credentials carry fixed lifetimes so a compromised key ages out, and a device that reinstalls rebinds a fresh address to its account through the hosted directory. Revocation is the hardest part to do without a network, because the whole point of revocation is to broadcast bad news. A gossip-carried revocation list is not something the protocol has today, and we say so rather than imply it.

The trade is deliberate. A server-backed system can revoke instantly for anyone who can reach the server, and not at all for anyone who cannot. A gossip-based system revokes eventually for everyone, including the devices that never touch the internet during the incident. At the disconnected edge, eventual-for-everyone beats instant-for-the-connected, because the connected devices were never the ones at risk.

Identity as the base layer for coordination

Identity is the layer every other guarantee rests on. Encrypted sessions, service discovery and invocation, authenticated relay, all of it assumes you can name the party on the other end and hold them to it. Doing that with a live directory is easy and well understood. Doing it with nothing but the keys the participants already carry is what makes coordination possible in the places Offline Protocol is built for.

Read more on offline identity, or talk to us about verifying identities where there is no authority to ask.

Frequently asked questions

Do Offline Protocol devices need a certificate authority to verify each other?

No. An identity is a keypair generated on the device, with no registration step with a central authority. Two devices that have no path to any server can still establish who they are talking to, because the proof travels with the party making the claim.

Can a relay device in the mesh pretend to be the sender?

No. The signature the recipient checks is the sender's, not the last hop's, and relays forward ciphertext they cannot read. Intermediaries move bytes; they do not vouch for anyone.

What happens if a contact's key changes?

After the first session, the key is pinned to that counterpart. Any later session that presents a different key for the same counterpart is a loud, detectable event rather than a silent substitution.

Does Offline Protocol have a revocation list?

Not today. A gossip-carried revocation list is not something the protocol has. Instead, sessions re-key on demand, MLS key packages and issued credentials carry fixed lifetimes so a compromised key ages out, and a device that reinstalls rebinds a fresh address to its account through the hosted directory.

What encryption do Offline Protocol sessions use?

Sessions are built on Ed25519 identities and encrypted with MLS (RFC 9420), the same primitives the transport layer uses end to end.

How does physical proximity help confirm a key?

Two people standing in the same room, or two vehicles in the same convoy, can confirm a key through a channel an attacker on the far side of the world cannot touch, which turns physical presence into a trust primitive.

More posts
Engineering Jul 23, 2026

Store, carry, forward: delivery when the path blinks in and out

Engineering Jul 14, 2026

Sending a message with no server