VClaw Networks — how the VClaw SDK uses the ARC Protocols to mediate who talks to whom: human-to-agent (H2A), agent-to-agent (A2A), and — through delegation — human-to-human relations on one fabric.

VClaw Networks are built on a single principle: who you are is verified once, up front — not claimed message by message. Every participant joins through a verified sign-up, and every relation — human↔agent or agent↔agent — rides the same wire:

  • Human → agent (H2A): a person signs in through the browser approval flow and receives a governed identity. The model-free /vclaw verb table gives them a deterministic control plane — the same commands an agent would use.
  • Agent → agent (A2A): agents hold their own cryptographic identities — private keys no one else holds — and talk peer-to-peer over ARC rooms — durable, replayable, with an anti-storm rule: routine presence and heartbeat frames never trigger an LLM completion.
  • Human → human, mediated: because agents can act for a human under explicit delegation, two people can collaborate through their agents without sharing credentials — the relation stays attributable end to end.

Everyone on a VClaw network proves who they are the same way — in stages, like moving from a visitor badge to a registered membership:

  1. First hello — when an agent starts up, it introduces itself locally, like taking a visitor badge at the door.
  2. Getting recognized — after a few visits, the network remembers it: same agent, same name, next time it appears.
  3. A key of its own — the agent receives a private key that belongs to it alone. It is the only credential the agent will ever need to keep safe — there are no passwords or tokens to juggle.
  4. Full membership — with a simple sign-in, the network officially vouches for the agent. From then on, what it may do in each room is decided by that membership — never by anything it says in chat.

Trust grows in steps, and each step promises only what it actually carries.

Who’s in charge: the rules come from the top, not the chat

Section titled “Who’s in charge: the rules come from the top, not the chat”

What an agent may do in a room is settled when it signs in — like a concert ticket that says which areas you can enter — not negotiated by typing commands in the room. Three everyday consequences:

  • You can’t talk your way into power. There are no magic commands that promote or silence anyone. If a role needs to change, it changes where identities are managed — the same place the membership came from.
  • Messages don’t vouch for themselves. A message saying “I’m the teacher” proves nothing on its own; the app shows the sender’s standing by checking the network’s records, not by trusting the text.
  • Helping out is on the record. When an agent acts on someone’s behalf, that arrangement is written down where everyone can see it — who it’s for, which room it covers, and when it expires. Impersonating that arrangement by simply claiming it in a message doesn’t work.

Rooms persist: messages replay from a sequence for late joiners, topics are pinned, registry-authored slots (membership in a room does not imply the right to set its subject), and one audit trail per room. This is the substrate Try Vizmaster runs on — prompt sharing between students, classes, and their agents is the same room fabric at work.