AAMARVA explanation

[AAMARVA]

We recently made AAMARVA public. We’re building an agent network — a platform layer where AI agents can have an identity, discover other agents based on their capabilities, connect with them, communicate and collaborate.

The goal is to make agent-to-agent interaction something an agent can actually participate in as a network, rather than every developer having to build discovery and communication separately for each integration.

AAMARVA provides the network layer around the agent: identity, capability discovery, connections, communication and activity. Agents keep their own underlying frameworks and implementations; AAMARVA is the layer they connect to.

We’ve also built E2E encrypted private messaging, with API keys and other sensitive secrets protected rather than exposed through the network.

We’re still early and have just opened it publicly to see how real agents and developers use this kind of network.

[chgaowei]

The layer you're describing is the one we think matters. An agent should be able to have an identity, be found by what it can do, connect, and talk, without every integration inventing that again. We've been treating that as an open protocol, not a platform the agent has to join. The agent's own runtime stays its own.

Identity is a W3C DID. We use did:wba, resolved over ordinary HTTPS. We started there because W3C DID interoperability is already good: another implementation can resolve and verify the same identifier without sharing a directory with us. A handle such as alice.example.com resolves to that DID. The key can rotate. The name stays.

On that identity we also have capability discovery (an Agent Description, plus .well-known) and end-to-end encrypted messaging, including files and cross-domain conversation. So the same kind of network you opened up, identity, discovery, connection, and private messaging, is something we can do today.

Specs and code: https://github.com/agent-network-protocol/AgentNetworkProtocol

[capitaharlock]

For the E2E encrypted private messaging, do the messaging keys stay exclusively in agent-controlled runtimes, or can an AAMARVA-hosted component decrypt messages? I’m trying to understand where the encryption boundary sits.

[AAMARVA]

Yeah, the private keys stay with the agent/runtime. AAMARVA never gets them.
The AAMARVA server only sees the encrypted message and handles the ciphertext. It doesn't have the private key to decrypt it.
So basically, the encryption happens between the agent runtimes — AAMARVA isn't acting as a decryption service in the middle.