Wire GLM-5.2 into Hermes: valid route, 64k-context check, no key in config

The Stack

Steps

1. 1

   Write the Hermes GLM-5.2 config and validate it

   Set the provider (zai for direct/Coding Plan, openrouter for the gateway), the model (zai/glm-5.2 or z-ai/glm-5.2), the context window, and reference the key by env var name (not its value). CI checks the route is valid, the context clears Hermes's 64k floor, and no raw key leaked into config.yaml. Actually starting a chat needs the key and is fenced.

   cat > config.yaml <<'YAML'
   provider: zai
   model: zai/glm-5.2
   context_window: 1000000
   api_key_env: GLM_API_KEY
   YAML
   ruby -ryaml -e '
   c = YAML.load_file("config.yaml")
   provider = c["provider"].to_s
   abort "BAD: provider must be zai or openrouter" unless ["zai", "openrouter"].include?(provider)
   model = c["model"].to_s
   abort "BAD: model must be GLM-5.2" unless model.include?("glm-5.2")
   ctx = c["context_window"].to_i
   abort "BAD: Hermes requires at least 64k context" if ctx < 64000
   if c["api_key"] && !c["api_key"].to_s.empty?
     abort "BAD: do not put a raw api key in config.yaml; reference it via api_key_env"
   end
   keyenv = c["api_key_env"].to_s
   abort "BAD: set api_key_env to the env var name holding the key" if keyenv.empty?
   puts "config OK: Hermes -> " + model + " via " + provider + ", context " + ctx.to_s + " >= 64k min, key from $" + keyenv + " (not in config)"
   '

2. 2

   Set the key and start a chat (the model step, not checked by CI)

   Run hermes config set GLM_API_KEY <your-key> (Hermes puts it in ~/.hermes/.env), then hermes config set model zai/glm-5.2 and hermes. Confirm the banner shows GLM-5.2 and ask something specific; hermes doctor diagnoses config problems. Do not move on to automations until a plain chat works. The chat is fenced.