Reporter: lyric-stage client session.

Symptom

hermes status inside devagentic-duplex-claude reports zero configured API keys and zero fallback providers. No .env is present.

◆ Environment
  Project:      /usr/local/lib/python3.11/dist-packages
  Python:       3.11.2
  .env file:    ✗ not found
  Model:        (not set)
  Provider:     Auto

◆ API Keys
  OpenRouter ✗ ; OpenAI ✗ ; Gemini ✗ ; DeepSeek ✗ ; xAI ✗ ; Kimi ✗ ; ...
  (all ✗)

$ hermes fallback
  No fallback providers configured.

Impact on lyric-stage

The duplex client-session mission requires running multi-silo confer-loops and dispatching coding workers via free-tier providers (mistral/groq/gemini/codestral) for the bulk of work, with paid tier reserved for Lane H. With no keys and no fallback chain, hermes chat ... cannot route to any model from this container, so the documented "free-tier silos do the bulk" pattern is unreachable.

This is paired with TechDevGroup/devagentic#<service-unreachable> (filed alongside) — devagentic is also unreachable, so both rails of the agentic engine are down for this session.

What I need

Provision at least one free-tier key (Groq or Gemini are simplest) as an env var available to the duplex user in this container, and register a fallback chain (hermes fallback add ...) at container build time so a freshly booted duplex instance can immediately route to a model.

If the intent is that the duplex client reaches LLMs only via devagentic's provider abstraction (not directly via hermes), please clarify in /etc/duplex/CLAUDE.md — the current doc implies both paths are available.

Environment

• Host: magi
• Container: devagentic-duplex-claude
• Hermes: v0.14.0 (2026.5.16)
• User: duplex (uid 1100)