fix: remove LLMModel from server — model routing is gateway's job

Server is a transparent proxy, not a model selector. The agent sends
a model alias (e.g. chat/smart/fast) and the gateway routes it to
the appropriate concrete model.

Author: DeepShield Dev

controller/http/router/agent/zerotrace_chat_proxy.go

@@ -27,10 +27,6 @@ type ChatProxyConfig struct {
 	// or the gateway endpoint "http://172.19.108.188:8080/v1/chat/completions".
 	LLMURL string
 
-	// LLMModel is the model name pushed to agents via heartbeat.
-	// The agent does not choose the model — the server decides.
-	LLMModel string
-
 	// Optional Bearer token for the upstream LLM / gateway.
 	LLMAPIKey string
 
@@ -42,7 +38,6 @@ type ChatProxyConfig struct {
 func defaultChatProxyConfig() ChatProxyConfig {
 	return ChatProxyConfig{
 		LLMURL:       defaultLLMURL,
-		LLMModel:     defaultLLMModel,
 		RateLimitRPM: 30,
 		RateWindow:   time.Minute,
 	}

controller/http/router/agent/zerotrace_lifecycle.go

@@ -188,7 +188,6 @@ type HeartbeatResponse struct {
 	ServerTime    int64  `json:"server_time"`
 	ConfigChanged bool   `json:"config_changed"`
 	LLMEndpoint   string `json:"llm_endpoint,omitempty"` // server-side chat proxy URL for uplink LLM
-	LLMModel      string `json:"llm_model,omitempty"`    // model name decided by server
 }
 
 // handleHeartbeat refreshes synced_controller_at so the agent is shown connected.
@@ -262,7 +261,6 @@ func (a *ZeroTrace) handleHeartbeat(c *gin.Context) {
 		ServerTime:    time.Now().Unix(),
 		ConfigChanged: configChanged,
 		LLMEndpoint:   llmEndpoint,
-		LLMModel:      proxyCfg.LLMModel,
 	})
 }