fix(codex): scope 24h retention to Bedrock Mantle

agent/auxiliary_client.py

@@ -1060,7 +1060,7 @@ class _CodexCompletionsAdapter:
         try:
             from agent.transports.codex import (
                 _content_cache_key,
-                _prompt_cache_retention_for_model,
+                _default_prompt_cache_retention_for_request,
             )
             from utils import base_url_host_matches
 
@@ -1070,21 +1070,15 @@ class _CodexCompletionsAdapter:
                 base_url_host_matches(_host_src, "githubcopilot.com")
                 or base_url_host_matches(_host_src, "models.github.ai")
             )
-            _is_codex_backend = (
-                base_url_host_matches(_host_src, "chatgpt.com")
-                and "/backend-api/codex" in _host_src.lower()
-            )
             if not _is_xai and not _is_github and "prompt_cache_key" not in resp_kwargs:
                 _cache_key = _content_cache_key(instructions, resp_kwargs.get("tools"))
                 if _cache_key:
                     resp_kwargs["prompt_cache_key"] = _cache_key
-            if (
-                not _is_xai
-                and not _is_github
-                and not _is_codex_backend
-                and "prompt_cache_retention" not in resp_kwargs
-            ):
-                _cache_retention = _prompt_cache_retention_for_model(model)
+            if "prompt_cache_retention" not in resp_kwargs:
+                _cache_retention = _default_prompt_cache_retention_for_request(
+                    model,
+                    _host_src,
+                )
                 if _cache_retention:
                     resp_kwargs["prompt_cache_retention"] = _cache_retention
         except Exception:

agent/chat_completion_helpers.py

@@ -1076,6 +1076,7 @@ def build_api_kwargs(agent, api_messages: list) -> dict:
             tools=tools_for_api,
             reasoning_config=agent.reasoning_config,
             session_id=getattr(agent, "session_id", None),
+            base_url=agent.base_url,
             max_tokens=agent.max_tokens,
             timeout=agent._resolved_api_call_timeout(),
             request_overrides=agent.request_overrides,

agent/transports/codex.py

@@ -48,8 +48,23 @@ _EXTENDED_PROMPT_CACHE_MODEL_RE = re.compile(
 )
 
 
-def _prompt_cache_retention_for_model(model: str) -> Optional[str]:
-    """Return ``24h`` for models documented to support extended retention."""
+def _default_prompt_cache_retention_for_request(
+    model: str,
+    base_url: Any,
+) -> Optional[str]:
+    """Return ``24h`` for supported models on Amazon Bedrock Mantle."""
+    from utils import base_url_hostname
+
+    hostname_parts = base_url_hostname(str(base_url or "")).split(".")
+    is_bedrock_mantle = (
+        len(hostname_parts) == 4
+        and hostname_parts[0] == "bedrock-mantle"
+        and bool(hostname_parts[1])
+        and hostname_parts[2:] == ["api", "aws"]
+    )
+    if not is_bedrock_mantle:
+        return None
+
     normalized = str(model or "").strip().lower().replace("_", "-")
     if _EXTENDED_PROMPT_CACHE_MODEL_RE.search(normalized):
         return "24h"
@@ -313,13 +328,11 @@ class ResponsesApiTransport(ProviderTransport):
         if not is_github_responses and not is_xai_responses and cache_key:
             kwargs["prompt_cache_key"] = cache_key
 
-        cache_retention = _prompt_cache_retention_for_model(model)
-        if (
-            cache_retention
-            and not is_github_responses
-            and not is_xai_responses
-            and not is_codex_backend
-        ):
+        cache_retention = _default_prompt_cache_retention_for_request(
+            model,
+            params.get("base_url"),
+        )
+        if cache_retention:
             kwargs.setdefault("prompt_cache_retention", cache_retention)
 
         if reasoning_enabled and is_xai_responses:

tests/agent/test_auxiliary_client.py

@@ -4965,7 +4965,7 @@ class TestCodexAdapterPromptCacheKey:
     ])
     def test_extended_cache_models_set_prompt_cache_retention(self, model):
         adapter, captured = self._build_adapter(
-            base_url="https://responses.example.com/v1",
+            base_url="https://bedrock-mantle.us-west-2.api.aws/v1",
             model=model,
         )
         adapter.create(messages=[
@@ -4982,6 +4982,19 @@ class TestCodexAdapterPromptCacheKey:
         ])
         assert "prompt_cache_retention" not in captured
 
+    @pytest.mark.parametrize("base_url", [
+        "https://api.openai.com/v1",
+        "https://example.services.ai.azure.com/openai/v1",
+        "https://responses.example.com/v1",
+    ])
+    def test_prompt_cache_retention_skipped_for_other_compatible_endpoints(self, base_url):
+        adapter, captured = self._build_adapter(base_url=base_url)
+        adapter.create(messages=[
+            {"role": "system", "content": "SYS"},
+            {"role": "user", "content": "hi"},
+        ])
+        assert "prompt_cache_retention" not in captured
+
     def test_prompt_cache_retention_skipped_for_xai_and_github_hosts(self):
         adapter, captured = self._build_adapter(base_url="https://api.x.ai/v1")
         adapter.create(messages=[

tests/agent/transports/test_codex_transport.py

@@ -264,6 +264,7 @@ class TestCodexBuildKwargs:
         kw = transport.build_kwargs(
             model=model, messages=messages, tools=[],
             session_id="test-session",
+            base_url="https://bedrock-mantle.us-west-2.api.aws/v1",
         )
         assert kw["prompt_cache_retention"] == "24h"
 
@@ -274,20 +275,29 @@ class TestCodexBuildKwargs:
             messages=[{"role": "user", "content": "Hi"}],
             tools=[],
             session_id="test-session",
+            base_url="https://bedrock-mantle.us-west-2.api.aws/v1",
         )
         assert "prompt_cache_retention" not in kw
 
-    def test_prompt_cache_retention_skipped_for_known_incompatible_backends(self, transport):
-        messages = [{"role": "user", "content": "Hi"}]
-        assert "prompt_cache_retention" not in transport.build_kwargs(
-            model="gpt-5.4", messages=messages, tools=[], is_xai_responses=True
-        )
-        assert "prompt_cache_retention" not in transport.build_kwargs(
-            model="gpt-5.4", messages=messages, tools=[], is_github_responses=True
-        )
-        assert "prompt_cache_retention" not in transport.build_kwargs(
-            model="gpt-5.4", messages=messages, tools=[], is_codex_backend=True
+    @pytest.mark.parametrize("base_url", [
+        "https://api.openai.com/v1",
+        "https://example.openai.azure.com/openai/v1",
+        "https://api.x.ai/v1",
+        "https://models.github.ai/inference",
+        "https://api.githubcopilot.com",
+        "https://chatgpt.com/backend-api/codex",
+        "https://responses.example.com/v1",
+        "https://bedrock-mantle.us-west-2.api.aws.example/v1",
+        "https://example.com/bedrock-mantle.us-west-2.api.aws/v1",
+    ])
+    def test_prompt_cache_retention_omitted_for_non_mantle_endpoints(self, transport, base_url):
+        kw = transport.build_kwargs(
+            model="gpt-5.4",
+            messages=[{"role": "user", "content": "Hi"}],
+            tools=[],
+            base_url=base_url,
         )
+        assert "prompt_cache_retention" not in kw
 
     def test_xai_responses_sends_cache_key_via_extra_body(self, transport):
         """xAI's Responses API documents ``prompt_cache_key`` as the

tests/run_agent/test_run_agent_codex_responses.py

@@ -396,6 +396,25 @@ def test_build_api_kwargs_codex(monkeypatch):
     assert "extra_body" not in kwargs
 
 
+def test_build_api_kwargs_mantle_sets_extended_prompt_cache_retention(monkeypatch):
+    _patch_agent_bootstrap(monkeypatch)
+    agent = run_agent.AIAgent(
+        model="openai.gpt-5.5",
+        provider="custom",
+        api_mode="codex_responses",
+        base_url="https://bedrock-mantle.us-west-2.api.aws/v1",
+        api_key="test-token",
+        quiet_mode=True,
+        max_iterations=1,
+        skip_context_files=True,
+        skip_memory=True,
+    )
+
+    kwargs = agent._build_api_kwargs([{"role": "user", "content": "Ping"}])
+
+    assert kwargs["prompt_cache_retention"] == "24h"
+
+
 def test_build_api_kwargs_codex_clamps_minimal_effort(monkeypatch):
     """'minimal' reasoning effort is clamped to 'low' on the Responses API.