fix(agents): preserve memory when summarization fails

camel/agents/chat_agent.py

@@ -1046,10 +1046,11 @@ def _get_context_with_summarization(
                 f"exceed limit, full compression."
             )
             summary = self.summarize(include_summaries=True)
-            self._update_memory_with_summary(
-                summary.get("summary", ""), include_summaries=True
-            )
-            return self.memory.get_context()
+            if self._try_update_memory_with_summary(
+                summary, include_summaries=True
+            ):
+                return self.memory.get_context()
+            return openai_messages, num_tokens
 
         threshold = self._calculate_next_summary_threshold()
         if num_tokens > threshold:
@@ -1058,10 +1059,11 @@ def _get_context_with_summarization(
                 f"({threshold}). Triggering summarization."
             )
             summary = self.summarize(include_summaries=False)
-            self._update_memory_with_summary(
-                summary.get("summary", ""), include_summaries=False
-            )
-            return self.memory.get_context()
+            if self._try_update_memory_with_summary(
+                summary, include_summaries=False
+            ):
+                return self.memory.get_context()
+            return openai_messages, num_tokens
 
         return openai_messages, num_tokens
 
@@ -1082,10 +1084,11 @@ async def _get_context_with_summarization_async(
                 f"exceed limit, full compression."
             )
             summary = await self.asummarize(include_summaries=True)
-            self._update_memory_with_summary(
-                summary.get("summary", ""), include_summaries=True
-            )
-            return self.memory.get_context()
+            if self._try_update_memory_with_summary(
+                summary, include_summaries=True
+            ):
+                return self.memory.get_context()
+            return openai_messages, num_tokens
 
         threshold = self._calculate_next_summary_threshold()
         if num_tokens > threshold:
@@ -1094,10 +1097,11 @@ async def _get_context_with_summarization_async(
                 f"({threshold}). Triggering summarization."
             )
             summary = await self.asummarize(include_summaries=False)
-            self._update_memory_with_summary(
-                summary.get("summary", ""), include_summaries=False
-            )
-            return self.memory.get_context()
+            if self._try_update_memory_with_summary(
+                summary, include_summaries=False
+            ):
+                return self.memory.get_context()
+            return openai_messages, num_tokens
 
         return openai_messages, num_tokens
 
@@ -1143,6 +1147,23 @@ def _calculate_next_summary_threshold(self) -> int:
 
         return threshold
 
+    def _try_update_memory_with_summary(
+        self,
+        summary_result: Dict[str, Any],
+        include_summaries: bool = False,
+    ) -> bool:
+        r"""Update memory when summary text exists, even if saving failed."""
+        summary = summary_result.get("summary", "")
+        if not isinstance(summary, str) or not summary.strip():
+            logger.warning(
+                "Skipping memory update because summarization failed: %s",
+                summary_result.get("status") or "unknown error",
+            )
+            return False
+
+        self._update_memory_with_summary(summary, include_summaries)
+        return True
+
     def _update_memory_with_summary(
         self, summary: str, include_summaries: bool = False
     ) -> None:

test/agents/test_chat_agent.py

@@ -292,6 +292,114 @@ def test_clean_snapshot_in_memory_skips_missing_records():
     assert entry.cached is True
 
 
+def _make_agent_with_summarization_history(token_limit=100_000):
+    model = DummyModel(ModelType.GPT_4O_MINI)
+    model._token_counter = MagicMock()
+    model._token_counter.count_tokens_from_messages.return_value = 10
+    agent = ChatAgent(
+        system_message="You are a helpful assistant.",
+        model=model,
+        token_limit=token_limit,
+        summarize_threshold=50,
+    )
+    agent.update_memory(
+        BaseMessage.make_user_message("user", "Remember this."),
+        OpenAIBackendRole.USER,
+    )
+    agent.update_memory(
+        BaseMessage.make_assistant_message("assistant", "I will."),
+        OpenAIBackendRole.ASSISTANT,
+    )
+    agent._calculate_next_summary_threshold = MagicMock(return_value=0)
+    return agent
+
+
+@pytest.mark.parametrize(
+    "summary_result",
+    [
+        {
+            "summary": "",
+            "status": "Failed to generate summary using model: transient",
+        },
+        {"summary": "", "status": "success"},
+    ],
+)
+def test_failed_automatic_summarization_preserves_memory(summary_result):
+    agent = _make_agent_with_summarization_history()
+    context_before = agent.memory.get_context()
+    agent.summarize = MagicMock(return_value=summary_result)
+
+    context_after = agent._get_context_with_summarization()
+
+    assert context_after == context_before
+    assert agent.memory.get_context() == context_before
+
+
+@pytest.mark.asyncio
+async def test_failed_async_automatic_summarization_preserves_memory():
+    agent = _make_agent_with_summarization_history()
+    context_before = agent.memory.get_context()
+    agent.asummarize = AsyncMock(
+        return_value={
+            "summary": "",
+            "status": "Failed to generate summary from model response.",
+        }
+    )
+
+    context_after = await agent._get_context_with_summarization_async()
+
+    assert context_after == context_before
+    assert agent.memory.get_context() == context_before
+
+
+@pytest.mark.asyncio
+@pytest.mark.parametrize("async_mode", [False, True])
+@pytest.mark.parametrize("include_summaries", [False, True])
+async def test_automatic_summarization_uses_summary_when_save_fails(
+    async_mode, include_summaries
+):
+    agent = _make_agent_with_summarization_history(token_limit=1000)
+    count_tokens = agent.model_backend.token_counter.count_tokens_from_messages
+    count_tokens.side_effect = lambda messages: 30 * len(messages)
+    # Exercise the real threshold so repeated context builds detect retries.
+    del agent._calculate_next_summary_threshold
+    for _ in range(10):
+        agent.update_memory(
+            BaseMessage.make_user_message("user", "Remember this."),
+            OpenAIBackendRole.USER,
+        )
+        agent.update_memory(
+            BaseMessage.make_assistant_message("assistant", "I will."),
+            OpenAIBackendRole.ASSISTANT,
+        )
+    if include_summaries:
+        agent._summary_token_count = agent.token_limit
+    summary = "[CONTEXT_SUMMARY] A perfectly good summary."
+    result = {"summary": summary, "status": "Error: write failed"}
+    summarizer = (
+        AsyncMock(return_value=result)
+        if async_mode
+        else MagicMock(return_value=result)
+    )
+    if async_mode:
+        agent.asummarize = summarizer
+    else:
+        agent.summarize = summarizer
+
+    for _ in range(5):
+        if async_mode:
+            messages, _ = await agent._get_context_with_summarization_async()
+        else:
+            messages, _ = agent._get_context_with_summarization()
+        assert len(messages) == 3
+        assert messages[1]["content"] == summary
+        assert "Remember this." in messages[2]["content"]
+
+    summarizer.assert_called_once_with(include_summaries=include_summaries)
+    if async_mode:
+        summarizer.assert_awaited_once()
+
+
 @pytest.mark.model_backend
 def test_chat_agent_stored_messages():
     system_msg = BaseMessage(