feat: add LM Studio JIT load mode

agent/agent_init.py

@@ -1354,6 +1354,27 @@ def init_agent(
         _agent_cfg = _load_agent_config()
     except Exception:
         _agent_cfg = {}
+
+    # LM Studio can either be explicitly preloaded through LM Studio's
+    # management API (the historical Hermes behavior) or left to LM Studio's
+    # just-in-time / Auto-Evict chat-completions path.  Keep the default
+    # explicit for backward compatibility; users with LM Studio Auto-Evict can
+    # opt into JIT via ``model.lmstudio_load_mode: jit``.
+    agent.lmstudio_load_mode = "explicit"
+    try:
+        _model_section = _agent_cfg.get("model", {})
+        if isinstance(_model_section, dict):
+            _load_mode = str(_model_section.get("lmstudio_load_mode", "explicit") or "explicit").strip().lower()
+            if _load_mode in {"explicit", "jit"}:
+                agent.lmstudio_load_mode = _load_mode
+            else:
+                logger.warning(
+                    "Invalid model.lmstudio_load_mode=%r; expected 'explicit' or 'jit'. Using explicit.",
+                    _model_section.get("lmstudio_load_mode"),
+                )
+    except Exception:
+        agent.lmstudio_load_mode = "explicit"
+
     try:
         agent._tool_guardrails = ToolCallGuardrailController(
             ToolCallGuardrailConfig.from_mapping(

run_agent.py

@@ -770,10 +770,13 @@ class AIAgent:
 
     def _ensure_lmstudio_runtime_loaded(self, config_context_length: Optional[int] = None) -> None:
         """
-        Preload the LM Studio model with at least Hermes' minimum context.
+        Preload the LM Studio model unless configured to rely on LM Studio JIT loading.
         """
         if (self.provider or "").strip().lower() != "lmstudio":
             return
+        if (getattr(self, "lmstudio_load_mode", "explicit") or "explicit").strip().lower() == "jit":
+            logger.debug("LM Studio explicit preload skipped: lmstudio_load_mode=jit")
+            return
         try:
             from agent.model_metadata import MINIMUM_CONTEXT_LENGTH
             from hermes_cli.models import ensure_lmstudio_model_loaded

tests/run_agent/test_lmstudio_load_mode.py

@@ -0,0 +1,63 @@
+from types import SimpleNamespace
+from typing import Any, cast
+
+from run_agent import AIAgent
+
+
+def _agent(load_mode="explicit"):
+    return SimpleNamespace(
+        provider="lmstudio",
+        model="test/model",
+        base_url="http://127.0.0.1:1234/v1",
+        api_key="",
+        lmstudio_load_mode=load_mode,
+        _config_context_length=None,
+        context_compressor=None,
+        api_mode="chat_completions",
+    )
+
+
+def test_lmstudio_jit_load_mode_skips_explicit_preload(monkeypatch):
+    calls = []
+
+    def fake_ensure(*args, **kwargs):
+        calls.append((args, kwargs))
+        return 64000
+
+    monkeypatch.setattr("hermes_cli.models.ensure_lmstudio_model_loaded", fake_ensure)
+
+    AIAgent._ensure_lmstudio_runtime_loaded(cast(Any, _agent("jit")))
+
+    assert calls == []
+
+
+def test_lmstudio_explicit_load_mode_preserves_preload(monkeypatch):
+    calls = []
+
+    def fake_ensure(*args, **kwargs):
+        calls.append((args, kwargs))
+        return 64000
+
+    monkeypatch.setattr("hermes_cli.models.ensure_lmstudio_model_loaded", fake_ensure)
+
+    AIAgent._ensure_lmstudio_runtime_loaded(cast(Any, _agent("explicit")))
+
+    assert len(calls) == 1
+    assert calls[0][0][:3] == ("test/model", "http://127.0.0.1:1234/v1", "")
+    assert calls[0][0][3] == 64000
+
+
+def test_missing_lmstudio_load_mode_defaults_to_explicit(monkeypatch):
+    calls = []
+    agent = _agent()
+    delattr(agent, "lmstudio_load_mode")
+
+    def fake_ensure(*args, **kwargs):
+        calls.append((args, kwargs))
+        return 64000
+
+    monkeypatch.setattr("hermes_cli.models.ensure_lmstudio_model_loaded", fake_ensure)
+
+    AIAgent._ensure_lmstudio_runtime_loaded(cast(Any, agent))
+
+    assert len(calls) == 1

website/docs/integrations/providers.md

@@ -844,7 +844,7 @@ hermes model
 # If LM Studio server auth is enabled, enter LM_API_KEY when prompted
 ```
 
-Hermes will automatically load a LM Studio model with 64K context length
+By default, Hermes explicitly asks LM Studio to load the selected model with 64K context length before the first request.
 
 To change context length in LM Studio:
 
@@ -860,6 +860,18 @@ You can use the CLI to estimate if the model will fit: `lms load model-name --co
 To set persistent per-model defaults: My Models tab → gear icon on the model → set context size.
 :::
 
+If you use LM Studio's Just-In-Time loading / Auto-Evict feature and want LM Studio to manage model loading and eviction from normal chat requests, skip Hermes' explicit preload step:
+
+```bash
+hermes config set model.lmstudio_load_mode jit
+```
+
+Set it back to the default explicit preload behavior with:
+
+```bash
+hermes config set model.lmstudio_load_mode explicit
+```
+
 **Tool calling:** Supported since LM Studio 0.3.6. Models with native tool-calling training (Qwen 2.5, Llama 3.x, Mistral, Hermes) are auto-detected and shown with a tool badge. Other models use a generic fallback that may be less reliable.
 
 ---