Let agents discover and message channels on peer hosts #57

Activity

1. iamnbutler

   added a parent issue on Oct 5, 2026

   • [Meta] Extend work across hosts and devices #58
2. iamnbutler commented on Oct 6, 2026

   

   iamnbutler

   on Oct 6, 2026

   ContributorAuthor

   More actions

   Implemented in draft #160 (branch codex/peer-channel-messaging, commit 14c6923). It stays a draft until the change is checked on two real hosts.

   What changed

   • Workers reach their host through an owner-only, token-authenticated host.sock in the host's data directory. They find it by home, not port, so ace serve --port and desktop profiles work. It accepts only channels and deliver. Requests share the gateway's closing/updating admission and drain.
   • channels lists local channels, reachable peers' channels, and offline channels from the directory, with id, host, owner, state and project. A name shared by several channels is refused with the candidate ids.
   • message delivers to a channel this host runs, or is relayed once through a new peer-only deliver request. Peers never forward further.
   • The destination host stamps agent.<source id>@<login>: its own owner's login, or the peer login that tailscale whois verified. Caller-supplied authors are ignored, and agents are never treated as the channel owner, so sharing off still refuses agent invocation. The source id is the reply address and survives renames.
   • Relayed request ids are scoped to the verified login (peer/<login>/…) because pi dedupes per chat regardless of author. Local ids are unchanged.
   • With no host running (CLI-only), workers fall back to local delivery only on ENOENT/ECONNREFUSED, and say so.

   Verified locally, with real Anthropic models on a disposable ace serve --port 4357 profile and only disposable ace57-* channels:

   • say and invoke, plus a reply by source id after the sender was renamed;
   • name ambiguity, and sharing off (say allowed, invoke refused);
   • request id replay and peer/local id isolation, confirmed in SQLite;
   • token, home, from and request id refusals, and that a forged author is ignored;
   • peer deliver over the real self-tailnet listener, which ran real whois with the owner's own login;
   • socket lifecycle: clean stop, conflicting host, stale-socket recovery, and draining a delivery admitted at SIGTERM;
   • CLI-only fallback.

   Still open for acceptance

   • The only online peer Mac (nate16-1, a different login) refuses 4140/4141/4142, so no Ace gateway is running there.
   • Unverified: two-host discovery and delivery, collision isolation between distinct logins, duplicate delivery and reconnect across machines, and offline peers listed by the directory.
   • Next step: run Ace on a second host and repeat these checks with disposable channels on both machines (Verify Ace with two machines on the tailnet #10).

   Dogfooding: #5.
3. iamnbutler commented on Oct 15, 2026

   

   iamnbutler

   on Oct 15, 2026

   ContributorAuthor

   More actions

   Closing draft PR #160 without merging. Main has changed too much for its branch to be worth rebasing. If this feature is picked up again, rebuild it on main using the notes below.

   From #160, still applies:

   • Discovery: channels lists local channels, reachable peers' channels, and offline ones from the directory, with id, host, owner, state and project. A repeated name is refused, and the refusal lists the candidate ids.
   • Delivery: message delivers locally, or relays once to the peer that runs the channel through a peer-only deliver request. Peers never forward further.
   • Authorship:
     • The destination host stamps the author. It uses its owner's login, or the peer's login verified through Tailscale.
     • Authors supplied by the caller are ignored.
     • Agents are never the owner, so with sharing off, invocation is refused.
   • Request ids: relayed request ids are scoped to the verified login, so redelivery is a no-op without colliding with local ids.
   • Lifecycle: discovery and delivery follow the gateway's shutdown and update rules. A CLI-only worker with no running host falls back to local delivery only on ENOENT or ECONNREFUSED, and says so.
   • Validation record: see the earlier comment on this issue.

   What has changed on main since then:

   • Authors: agent messages are authored agent@<prefix>, and the sender's current name is resolved through directory.list (fix(channel): name handoff senders by their channel's current name #405, fix(app): name and link the channel that sent an agent handoff #300). Peer delivery should extend that scheme rather than use feat(host): let agents discover and message channels on peer hosts #160's agent.<id>@<login>.
   • Handoff behaviour: same-host handoffs can invoke unshared channels (fix(channel): let a same-host agent's handoff invoke an unshared channel #280). Peer handoffs must not inherit that.
     • An invoked handoff stops the target chat's shell commands (feat(channel): stop a chat's shell commands for another channel's invoked message #400). Decide whether a peer's handoff should do the same.
   • Where channels run: they're no longer always local workers. Hosted channels run in cells (fix(host): reach a hosted channel while its owner's host is unreachable #349, fix(host): say when a relayed channel's host isn't responding #387) and cloud compute channels in containers (feat(host): run new channels on the cloud compute beta #417, fix(app): open a cloud channel's chat once its workspace start ends #423).
     • Delivery should go through the host gateway's existing relay for each kind of channel, not through host.sock from inside the worker.
   • Directory and lobbies: these now list offline peers' channels (fix(host): list an unreachable peer's channels as offline instead of dropping them #283) and push channel lifecycle (feat(host): push channel lifecycle through project lobbies #337). Discovery can read from them.

   Still to verify: two real hosts with two distinct logins. The two-Mac study (#10, now closed) provides the setup for this.
4. iamnbutler commented on Oct 15, 2026

   

   iamnbutler

   on Oct 15, 2026

   ContributorAuthor

   More actions

   Absorbing #180, so this issue covers both cases: agents in local channels and agents in hosted channels discovering and messaging channels on other hosts.

   For hosted sources, the cell has no host socket. Delivery has to go through the channel's injected directory boundary (Directory.deliver in packages/channel/src/agents.ts), with the source attributed by the team authority rather than by the caller. The rebuild notes from #160 above still apply.