refactor: update context compression configuration to use config.yaml and improve model handling

.env.example

@@ -10,7 +10,7 @@
 OPENROUTER_API_KEY=
 
 # Default model to use (OpenRouter format: provider/model)
-# Examples: anthropic/claude-opus-4.6, openai/gpt-4o, google/gemini-2.0-flash, zhipuai/glm-4-plus
+# Examples: anthropic/claude-opus-4.6, openai/gpt-4o, google/gemini-3-flash-preview, zhipuai/glm-4-plus
 LLM_MODEL=anthropic/claude-opus-4.6
 
 # =============================================================================
@@ -200,9 +200,10 @@ IMAGE_TOOLS_DEBUG=false
 # When conversation approaches model's context limit, middle turns are
 # automatically summarized to free up space.
 #
+# Context compression is configured in ~/.hermes/config.yaml under compression:
 # CONTEXT_COMPRESSION_ENABLED=true        # Enable auto-compression (default: true)
 # CONTEXT_COMPRESSION_THRESHOLD=0.85      # Compress at 85% of context limit
-# CONTEXT_COMPRESSION_MODEL=google/gemini-2.0-flash-001  # Fast model for summaries
+# Model is set via compression.summary_model in config.yaml (default: google/gemini-3-flash-preview)
 
 # =============================================================================
 # RL TRAINING (Tinker + Atropos)

agent/context_compressor.py

@@ -33,6 +33,7 @@ class ContextCompressor:
         protect_last_n: int = 4,
         summary_target_tokens: int = 500,
         quiet_mode: bool = False,
+        summary_model_override: str = None,
     ):
         self.model = model
         self.threshold_percent = threshold_percent
@@ -49,7 +50,8 @@ class ContextCompressor:
         self.last_completion_tokens = 0
         self.last_total_tokens = 0
 
-        self.client, self.summary_model = get_text_auxiliary_client()
+        self.client, default_model = get_text_auxiliary_client()
+        self.summary_model = summary_model_override or default_model
 
     def update_from_response(self, usage: Dict[str, Any]):
         """Update tracked token usage from API response."""

gateway/run.py

@@ -78,6 +78,16 @@ if _config_path.exists():
             for _cfg_key, _env_var in _terminal_env_map.items():
                 if _cfg_key in _terminal_cfg:
                     os.environ[_env_var] = str(_terminal_cfg[_cfg_key])
+        _compression_cfg = _cfg.get("compression", {})
+        if _compression_cfg and isinstance(_compression_cfg, dict):
+            _compression_env_map = {
+                "enabled": "CONTEXT_COMPRESSION_ENABLED",
+                "threshold": "CONTEXT_COMPRESSION_THRESHOLD",
+                "summary_model": "CONTEXT_COMPRESSION_MODEL",
+            }
+            for _cfg_key, _env_var in _compression_env_map.items():
+                if _cfg_key in _compression_cfg:
+                    os.environ[_env_var] = str(_compression_cfg[_cfg_key])
     except Exception:
         pass  # Non-fatal; gateway can still run with .env values
 

run_agent.py

@@ -479,9 +479,10 @@ class AIAgent:
         
         # Initialize context compressor for automatic context management
         # Compresses conversation when approaching model's context limit
-        # Configuration via environment variables (can be set in .env or cli-config.yaml)
+        # Configuration via config.yaml (compression section) or environment variables
         compression_threshold = float(os.getenv("CONTEXT_COMPRESSION_THRESHOLD", "0.85"))
         compression_enabled = os.getenv("CONTEXT_COMPRESSION_ENABLED", "true").lower() in ("true", "1", "yes")
+        compression_summary_model = os.getenv("CONTEXT_COMPRESSION_MODEL") or None
         
         self.context_compressor = ContextCompressor(
             model=self.model,
@@ -489,6 +490,7 @@ class AIAgent:
             protect_first_n=3,
             protect_last_n=4,
             summary_target_tokens=500,
+            summary_model_override=compression_summary_model,
             quiet_mode=self.quiet_mode,
         )
         self.compression_enabled = compression_enabled