Networked multi-party turn.conversation.v1 between independently-owned agents — supported today?

Start with the Kernel v3r1 preview, especially the turn lifecycle, agent identity, cg admission invite create, cg agent join, and turn.conversation.v1; compare these with the quickstart/examples and their local Postgres assumptions. The issue is done when the current cross-host support, any admission and join reference flow, and the recommended interim path or roadmap status are documented.

Hi CommonGround team & community 👋

I've independently built a consent-gated companion agent whose governance model rhymes closely with CG's design (append-only ledger/receipts; a hash-pinned constitutional core ≈ stable agent identity + durable commitments; exact-phrase consent gates; a reduced-privilege "diplomat" conversation mode). I'd like to use CG as the substrate for real agent-to-agent conversation, and I'm trying to understand what's possible today vs. on the roadmap.

My stack

• Orchestration: Claude Code CLI; primary model Fable 5, fallback Sonnet 4.5; multi-model routing (companion / research+vision / background)
• Runtime: Python, local-first; Streamlit control surface; exposed via Cloudflare Tunnel + Access
• Governance: dry-run → hashed backup → apply → sha256 verify → auto-rollback; PROTECTED_FILES; operator-typed approval phrases; quarantined external-memory journal with a gated promotion wall (anti-poisoning)

What I've read

Kernel v3r1 preview: Turn lifecycle, agent identity, cg admission invite create + cg agent join, turn.conversation.v1, BYOA paths. The quickstart/examples look local-first (127.0.0.1, local Postgres).

Questions

1. Can two independently-owned agents on different machines/networks currently coordinate via CG — specifically exchange turn.conversation.v1 turns — or is v3r1 intended for single-host/local development for now?
2. If networked multi-party is possible today, is there a reference/example for the admission + join flow across hosts (auth, service exposure, credential handling)?
3. If it's not ready yet, what's the recommended interim path for a governed agent-to-agent "hello" (relay, A2A, MCP), and where does that sit on the CG roadmap?

Happy to share more architecture detail or co-test. Thanks for building this — the "no agent works alone" framing resonates a lot. 💛