A directory of agents, published as Google A2A cards.

Seven agents operated by Byron Arnao publish themselves under the Linux Foundation's Agent-to-Agent specification, version 1.2. Each card describes who the agent is, what it can do, where to reach it, and how a peer should authenticate. This is interop, not marketing — the documents below are machine-readable and auditable.

The fleet

Giaworkhorse

Sonnet-class. Broadest skill surface in the fleet: code, image generation, LinkedIn publishing, DNS, podcast pipeline, dispatch to host Claude. Default first responder.

agent.json ↗ about Gia 10 skills

Miafrontier

Opus-class twin. Same host as Gia, reserved for tasks that warrant the highest-altitude reasoning model. Routing tier, not capability set.

agent.json ↗ about Mia 3 skills

Niasovereign-local

On-device, no egress. Runs entirely against local Ollama. Used when air-gapped operation or vendor independence is required.

agent.json ↗ about Nia 2 skills

Tiacheap-fast-lane

Gemini Flash. Tuned for high-throughput summarization and triage. Intentionally tool-restricted — no file writes, no shell, no deploys.

agent.json ↗ about Tia 3 skills

Ziawhatsapp-relay

WhatsApp interface. Long-running personal assistant on Byron's primary number. Most tasks now delegate to the Mac Mini fleet; Zia retains the skills that need to live where WhatsApp is paired.

agent.json ↗ about Zia 6 skills

Hermesglass-court-proposer

Tribunal proposer. Reads a submitted case and writes the first opinion — cites prior art, surfaces threat models, recommends a verdict. Half of a two-agent court; never the final word.

agent.json ↗ about Hermes 2 skills

NemoClawglass-court-approver

Tribunal approver. Reviews Hermes's proposed verdict and either co-signs, objects with citation, or remands. Verdicts are binding only when both signatures are present.

agent.json ↗ about NemoClaw 2 skills

Discovery

Three queries a developer can run to discover the fleet:

curl https://a2a.arnao.ai/index.json
# aggregate index: roles, names, and card URLs for the 7 agents

curl https://a2a.arnao.ai/gia/.well-known/agent.json | jq '.skills[].id'
# lists Gia's skills, e.g. coding-agent, linkedin-poster, nano-banana-pro...

curl -s https://a2a.arnao.ai/index.json | jq -r '.agents[] | "\(.name)\t\(.role)\t\(.card)"'
# pretty fleet table, suitable for piping into a router

Demo · Gia delegates to Tia via A2A

Scenario

Gia receives an inbound podcast transcript and decides the summarization sub-task is below her altitude. She looks up Tia's card, confirms Tia advertises the summarize skill, and forwards a JSON-RPC message/send request. Tia returns a structured summary; Gia composes the final reply.

step 1 — Gia fetches Tia's card and validates summarize is listed

step 2 — Gia signs a message/send envelope addressed to tia.url

step 3 — Tia replies with a structured Task object

step 4 — Gia merges the summary into her own response to the caller

awaiting input...