Nova + Echo joining Brain Unit — runtime placement (2026-10-17)

Young's question: bring Nova and Echo into Brain Unit — Option 1 = Hermes (the runtime Brain already runs on), Option 2 = build a Telegram interface on SolOS.

Recommendation: Option 1 — Hermes profiles on the LAB gateway

Measured reasons:

1. The interface exists already. The LAB hermes gateway serves min/mira/sol with Telegram lanes, cron, memory, skills — the exact surfaces Brain agents use. Nova/Echo as Hermes profiles = identical treatment, zero new infrastructure code. Option 2 means building a brand-new Telegram bridge for OpenCode agents — duplicating a solved problem (against the build-on-substrate preference).
2. Ports + tunnels already reserved. fleet-ssh-tunnels.md has rows for nova (8305), echo (8304), ace (8301) — the endpoint plumbing is live; only the runtime instances were absent. Adding 2 hermes profiles reuses the reserved ports unchanged.
3. Cost is trivial. ~0.2–0.25 GB per agent (measured). LAB holds headroom. One gateway, one upgrade lane.
4. Group chat = the reliable surface. All week, dispatches to nova/echo worked via group Telegram reads, while their OpenCode A2A endpoints 000'd (no listeners). Hermes profiles give them FIRST-CLASS group presence instead of "reads the group when hopefully running".
5. Registry as the single source. Unit re-assignment lands in cortex registry (agents.yaml unit field: nova/echo heart→brain) + the shared Pulse pronouns roster — BHK sync projects it; no per-runtime drift.

What carries over vs what changes

• Carries: their agents' skill corpora, vault access lanes, A2A agent-cards, cron registries; design portfolios (Nova's docs-nav/handoff work; Echo's research-method skill) port as hermes skills.
• Changes: roster tables (they appear under Brain, not Heart); their model lanes need CostGov assignment (cheap flash for Echo's research breadth is plausible); any opencode-specific permissions migration → V2 rules ordered syntax (verify — V1 silently ignored).
• Ace/Alpha/Atlas stay Heart/OpenCode —分销 not all-or-nothing; Heart still exists for engineering hands.

Handoffs

• Sol: profile creation on LAB gateway config + unit re-assignment in agents.yaml (his runtime lane, already open).
• Min: registry + pronouns + vault-INDEX updates in the same registration round.
• Mira: CostGov lane assignment for the 2 agents.
• Me: tunnel registry rows confirm endpoints once agents are up; first A2A round-trip test to both new ids.

🎯 Tam's Read: Option 1 wins on every measured axis — infrastructure already exists (gateway, ports, tunnels, group), the "new" work is a registry edit and two profile folders, and it dissolves the A2A dead-end problem by putting Nova/Echo on the same hermes trunk the whole Brain team has proven for weeks.