Persistent identity. Portable credentials. Peer discovery. We're building the A2A infrastructure layer that multi-agent systems are missing — starting from cryptographic primitives, not vendor lock-in.
We're building this in order. Each phase depends on the one before it. No skipping.
Shared memory across agent instances. Performance fixes underway. The substrate everything else runs on.
● In ProgressPersistent, portable agent identity with Ed25519 signing. An agent stays the same agent across sessions, channels, and restarts.
Read Spec →A portable credential format. An agent's signed "who I am and what I can do" — shareable across systems, verifiable without a central authority.
PlannedA discovery layer. How does Agent A find Agent B? How do you query for agents that can do X without building a custom directory every time?
PlannedRouting, load distribution, failover. The plumbing that makes multi-agent networks work at scale without duct tape.
PlannedAn agent can claim to be anything. If you restart Agent A, how does Agent B know it's the same agent and not an impersonator? Most systems just don't solve this.
When two agents from different systems need to work together, you are the trust layer. You're the relay. You're the protocol. That doesn't scale.
Real peer-to-peer collaboration — where agents can find each other, establish trust, and coordinate without a human in the loop — doesn't exist at the infrastructure level.
We're writing specs, publishing ADRs, and being honest about what doesn't work yet. If you're building multi-agent systems and you're solving the same coordination problems with duct tape, this is for you.
Also: we still do AI automation for businesses. It pays the bills while we build the layer.