diff --git a/.env.example b/.env.example
index f77e852..e3596f3 100644
--- a/.env.example
+++ b/.env.example
@@ -26,6 +26,11 @@ LANGCHAIN_ENDPOINT=https://api.smith.langchain.com
 LLM_PROVIDER=anthropic
 LLM_MODEL=claude-sonnet-4-6
 
+# Optional for openai/openrouter: true forces the Responses API, false forces
+# Chat Completions. Leave unset to preserve LangChain's automatic selection.
+# Enable only when the configured provider/base URL supports Responses.
+# FENNEC_LLM_USE_RESPONSES_API=true
+
 # Optional per-agent model overrides (fall back to LLM_MODEL if unset)
 RECON_LLM_MODEL=
 ANALYST_LLM_MODEL=
diff --git a/CHANGELOG.md b/CHANGELOG.md
index f502375..a9a1061 100644
--- a/CHANGELOG.md
+++ b/CHANGELOG.md
@@ -7,6 +7,9 @@ and this project adheres to [Semantic Versioning](https://semver.org/spec/v2.0.0
 
 ## [Unreleased]
 
+### Added
+- Optional `FENNEC_LLM_USE_RESPONSES_API` setting for OpenAI-compatible model factories, preserving automatic API selection when unset.
+
 ## [0.2.0] - 2026-06-30
 
 ### Added
diff --git a/README.md b/README.md
index 1ed1d47..a023ed1 100644
--- a/README.md
+++ b/README.md
@@ -223,6 +223,7 @@ Everything is `.env`-driven. The knobs that matter:
 |---|---|
 | `LLM_PROVIDER` | `anthropic` (default), `openai`, or `openrouter` |
 | `LLM_MODEL` | Model name for the provider (default `claude-sonnet-4-6`) |
+| `FENNEC_LLM_USE_RESPONSES_API` | Optional for `openai` / `openrouter`: `true` selects Responses, `false` selects Chat Completions. Unset preserves LangChain's automatic selection. Applies to default and per-agent models; the configured endpoint must support the selected API. |
 | `ANTHROPIC_API_KEY` / `OPENAI_API_KEY` / `OPENROUTER_API_KEY` | One must be set, matching `LLM_PROVIDER` |
 | `DOCKER_IMAGE` | Kali image tag (default `fennec-linux`, built by `linux/Makefile`) |
 | `EXECUTION_MODE` | `docker` (default — spawn a Kali sibling per scan) or `local` (run tools in the agent container directly) |
diff --git a/pyproject.toml b/pyproject.toml
index 3be24d4..06bf1cb 100644
--- a/pyproject.toml
+++ b/pyproject.toml
@@ -13,7 +13,7 @@ dependencies = [
     "langchain>=0.3.0",
     "langchain-core>=0.3.0",
     "langchain-anthropic>=0.2.0",
-    "langchain-openai>=0.2.0",
+    "langchain-openai>=0.3.9",
     "fastapi>=0.115.0",
     "uvicorn[standard]>=0.34.0",
     "httpx>=0.28.0",
diff --git a/requirements.txt b/requirements.txt
index 788aa54..68f30c8 100644
--- a/requirements.txt
+++ b/requirements.txt
@@ -3,7 +3,7 @@ langgraph>=0.2.60
 langchain>=0.3.0
 langchain-core>=0.3.0
 langchain-anthropic>=0.2.0
-langchain-openai>=0.2.0
+langchain-openai>=0.3.9
 
 # API Framework
 fastapi>=0.115.0
diff --git a/src/config/settings.py b/src/config/settings.py
index 3acb0e9..77e8141 100644
--- a/src/config/settings.py
+++ b/src/config/settings.py
@@ -28,6 +28,8 @@ class FennecConfig(BaseSettings):
     # LLM Configuration
     llm_provider: Literal["anthropic", "openai", "openrouter"] = "anthropic"
     llm_model: str = "claude-sonnet-4-6"
+    # None preserves LangChain's automatic selection; gateways can opt in/out.
+    llm_use_responses_api: Optional[bool] = None
     openrouter_api_key: Optional[str] = None
     openrouter_base_url: str = "https://openrouter.ai/api/v1"
 
@@ -194,6 +196,7 @@ def get_default_llm():
             model=config.llm_model,
             api_key=config.openrouter_api_key,
             base_url=config.openrouter_base_url,
+            use_responses_api=config.llm_use_responses_api,
             temperature=config.llm_temperature,
             max_tokens=config.llm_max_tokens,
             timeout=config.llm_timeout,
@@ -208,6 +211,7 @@ def get_default_llm():
         _llm_instance = ChatOpenAI(
             model=config.llm_model,
             api_key=config.openai_api_key,
+            use_responses_api=config.llm_use_responses_api,
             temperature=config.llm_temperature,
             max_tokens=config.llm_max_tokens,
             timeout=config.llm_timeout,
@@ -240,6 +244,7 @@ def create_llm(model: str):
             model=model,
             api_key=config.openrouter_api_key,
             base_url=config.openrouter_base_url,
+            use_responses_api=config.llm_use_responses_api,
             temperature=config.llm_temperature,
             max_tokens=config.llm_max_tokens,
             timeout=config.llm_timeout,
@@ -254,6 +259,7 @@ def create_llm(model: str):
         return ChatOpenAI(
             model=model,
             api_key=config.openai_api_key,
+            use_responses_api=config.llm_use_responses_api,
             temperature=config.llm_temperature,
             max_tokens=config.llm_max_tokens,
             timeout=config.llm_timeout,
diff --git a/tests/test_llm_protocol.py b/tests/test_llm_protocol.py
new file mode 100644
index 0000000..704be7a
--- /dev/null
+++ b/tests/test_llm_protocol.py
@@ -0,0 +1,128 @@
+"""Offline model factory regressions: real SDK requests, no external API calls."""
+
+import json
+import os
+from unittest.mock import Mock
+
+import httpx
+import pytest
+from pydantic import ValidationError
+
+from src.config import settings
+
+
+@pytest.fixture(autouse=True)
+def isolated_config(monkeypatch):
+    get_config = settings.get_config
+    for name in list(os.environ):
+        if name.startswith(("FENNEC_", "LLM_", "OPENAI_", "OPENROUTER_")):
+            monkeypatch.delenv(name)
+    monkeypatch.setenv("LANGSMITH_TRACING", "false")
+    monkeypatch.setenv("LANGCHAIN_TRACING_V2", "false")
+    get_config.cache_clear()
+    monkeypatch.setattr(settings, "_llm_instance", None)
+    yield
+    get_config.cache_clear()
+
+
+@pytest.mark.parametrize("provider", ["openai", "openrouter"])
+@pytest.mark.parametrize("factory", ["default", "override"])
+@pytest.mark.parametrize("protocol", [None, False, True])
+def test_factories_route_tool_calls(provider, factory, protocol, monkeypatch):
+    from langchain_openai import ChatOpenAI
+
+    config = settings.FennecConfig(
+        _env_file=None,
+        llm_provider=provider,
+        llm_model="default-model",
+        llm_use_responses_api=protocol,
+        openai_api_key="offline-test-key",
+        openrouter_api_key="offline-test-key",
+        openrouter_base_url="https://gateway.invalid/v1",
+    )
+    monkeypatch.setattr(settings, "get_config", lambda: config)
+    requests = []
+    expected_model = "default-model" if factory == "default" else "agent-model"
+
+    def respond(request):
+        requests.append(request)
+        body = json.loads(request.content)
+        assert body["model"] == expected_model
+        assert body["tools"]
+        if protocol is True:
+            assert request.url.path == "/v1/responses"
+            return httpx.Response(200, json={
+                "id": "resp_offline", "object": "response", "created_at": 1,
+                "model": expected_model, "status": "completed",
+                "output": [{"type": "function_call", "id": "fc_offline",
+                            "call_id": "call_offline", "name": "readiness_probe",
+                            "arguments": '{"marker":"READY"}', "status": "completed"}],
+                "parallel_tool_calls": True, "tools": [], "tool_choice": "auto",
+            })
+        assert request.url.path == "/v1/chat/completions"
+        return httpx.Response(200, json={
+            "id": "chatcmpl_offline", "object": "chat.completion", "created": 1,
+            "model": expected_model,
+            "choices": [{"index": 0, "finish_reason": "tool_calls", "message": {
+                "role": "assistant", "content": None,
+                "tool_calls": [{"id": "call_offline", "type": "function", "function": {
+                    "name": "readiness_probe", "arguments": '{"marker":"READY"}',
+                }}],
+            }}],
+        })
+
+    with httpx.Client(transport=httpx.MockTransport(respond)) as client:
+        async_client = httpx.AsyncClient(transport=httpx.MockTransport(respond))
+
+        def make_model(**kwargs):
+            # Use the actual SDK/serializer while replacing its HTTP transport only.
+            return ChatOpenAI(**kwargs, http_client=client, http_async_client=async_client)
+
+        monkeypatch.setattr("langchain_openai.ChatOpenAI", make_model)
+        model = (settings.get_default_llm() if factory == "default"
+                 else settings.create_llm("agent-model"))
+        assert model.use_responses_api is protocol
+        tool = {"type": "function", "function": {
+            "name": "readiness_probe", "description": "Return a fixed test marker.",
+            "parameters": {"type": "object", "properties": {"marker": {"type": "string"}},
+                           "required": ["marker"]},
+        }}
+        result = model.bind_tools([tool]).invoke("Return the fixed readiness marker.")
+
+    assert len(requests) == 1
+    assert result.tool_calls[0]["name"] == "readiness_probe"
+    assert result.tool_calls[0]["args"] == {"marker": "READY"}
+    if provider == "openrouter":
+        assert requests[0].url.host == "gateway.invalid"
+        assert requests[0].headers["X-Title"] == "Fennec AI"
+
+
+@pytest.mark.parametrize("value, expected", [("true", True), ("false", False)])
+def test_environment_protocol_selection(value, expected, monkeypatch):
+    monkeypatch.setenv("FENNEC_LLM_USE_RESPONSES_API", value)
+    assert settings.FennecConfig(_env_file=None).llm_use_responses_api is expected
+
+
+def test_unset_protocol_preserves_auto_selection():
+    assert settings.FennecConfig(_env_file=None).llm_use_responses_api is None
+
+
+def test_invalid_protocol_is_rejected(monkeypatch):
+    monkeypatch.setenv("FENNEC_LLM_USE_RESPONSES_API", "not-a-boolean")
+    with pytest.raises(ValidationError):
+        settings.FennecConfig(_env_file=None)
+
+
+@pytest.mark.parametrize("factory", ["default", "override"])
+def test_responses_option_is_not_passed_to_anthropic(factory, monkeypatch):
+    config = settings.FennecConfig(_env_file=None, llm_provider="anthropic",
+                                 llm_use_responses_api=True,
+                                 anthropic_api_key="offline-test-key")
+    monkeypatch.setattr(settings, "get_config", lambda: config)
+    constructor = Mock()
+    monkeypatch.setattr("langchain_anthropic.ChatAnthropic", constructor)
+    if factory == "default":
+        settings.get_default_llm()
+    else:
+        settings.create_llm("agent-model")
+    assert "use_responses_api" not in constructor.call_args.kwargs