vectara-agentic 0.4.1__py3-none-any.whl → 0.4.3__py3-none-any.whl

tests/test_react_error_handling.py

@@ -0,0 +1,293 @@
+# Suppress external dependency warnings before any other imports
+import warnings
+
+warnings.simplefilter("ignore", DeprecationWarning)
+
+import unittest
+import asyncio
+
+from vectara_agentic.agent import Agent
+from vectara_agentic.tools import ToolsFactory
+from vectara_agentic.agent_config import AgentConfig
+from vectara_agentic.types import AgentType, ModelProvider
+
+import nest_asyncio
+
+nest_asyncio.apply()
+
+from conftest import (
+    AgentTestMixin,
+    react_config_anthropic,
+    react_config_gemini,
+    react_config_together,
+    mult,
+    STANDARD_TEST_TOPIC,
+    STANDARD_TEST_INSTRUCTIONS,
+    is_rate_limited,
+    is_api_key_error,
+)
+
+
+class TestReActErrorHandling(unittest.TestCase, AgentTestMixin):
+    """Test error handling and recovery for ReAct agents."""
+
+    def setUp(self):
+        self.tools = [ToolsFactory().create_tool(mult)]
+        self.topic = STANDARD_TEST_TOPIC
+        self.instructions = STANDARD_TEST_INSTRUCTIONS
+
+    def test_react_anthropic_rate_limit_handling(self):
+        """Test ReAct agent handling of Anthropic rate limits."""
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            response = agent.chat("What is 5 times 10?")
+            self.check_response_and_skip(response, "Anthropic")
+
+            # If we get a response, check it's valid
+            if response.response and not is_rate_limited(response.response):
+                self.assertIn("50", response.response)
+
+    def test_react_openai_error_handling(self):
+        """Test ReAct agent handling of OpenAI-specific errors."""
+        openai_react_config = AgentConfig(
+            agent_type=AgentType.REACT,
+            main_llm_provider=ModelProvider.OPENAI,
+            tool_llm_provider=ModelProvider.OPENAI,
+        )
+
+        agent = Agent(
+            agent_config=openai_react_config,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+        )
+
+        with self.with_provider_fallback("OpenAI"):
+            response = agent.chat("Calculate 7 times 8.")
+            self.check_response_and_skip(response, "OpenAI")
+
+            # If we get a response, check it's valid
+            if response.response and not is_rate_limited(response.response):
+                self.assertIn("56", response.response)
+
+    def test_react_together_error_handling(self):
+        """Test ReAct agent handling of Together.AI errors."""
+        agent = Agent(
+            agent_config=react_config_together,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+        )
+
+        with self.with_provider_fallback("Together AI"):
+            response = agent.chat("Calculate 4 times 15.")
+            self.check_response_and_skip(response, "Together AI")
+
+            # If we get a response, check it's valid
+            if response.response and not is_rate_limited(response.response):
+                self.assertIn("60", response.response)
+
+    def test_react_gemini_error_handling(self):
+        """Test ReAct agent handling of Gemini-specific errors."""
+        agent = Agent(
+            agent_config=react_config_gemini,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+        )
+
+        with self.with_provider_fallback("Gemini"):
+            response = agent.chat("What is 12 times 4?")
+            self.check_response_and_skip(response, "Gemini")
+
+            # If we get a response, check it's valid
+            if response.response and not is_rate_limited(response.response):
+                self.assertIn("48", response.response)
+
+    def test_react_async_error_handling(self):
+        """Test ReAct agent error handling during async operations."""
+        async def _async_test():
+            agent = Agent(
+                agent_config=react_config_anthropic,
+                tools=self.tools,
+                topic=self.topic,
+                custom_instructions=self.instructions,
+            )
+
+            with self.with_provider_fallback("Anthropic"):
+                # Test async chat error handling
+                response = await agent.achat("Calculate 11 times 3.")
+                self.check_response_and_skip(response, "Anthropic")
+
+                if response.response and not is_rate_limited(response.response):
+                    self.assertIn("33", response.response)
+
+        asyncio.run(_async_test())
+
+    def test_react_streaming_error_handling(self):
+        """Test ReAct agent error handling during streaming operations."""
+        async def _async_test():
+            agent = Agent(
+                agent_config=react_config_anthropic,
+                tools=self.tools,
+                topic=self.topic,
+                custom_instructions=self.instructions,
+            )
+
+            with self.with_provider_fallback("Anthropic"):
+                try:
+                    stream = await agent.astream_chat("Calculate 13 times 2.")
+
+                    # Consume the stream
+                    chunks = []
+                    async for chunk in stream.async_response_gen():
+                        chunks.append(str(chunk))
+
+                    response = await stream.aget_response()
+                    self.check_response_and_skip(response, "Anthropic")
+
+                    if response.response and not is_rate_limited(response.response):
+                        self.assertIn("26", response.response)
+
+                except Exception as e:
+                    error_msg = str(e)
+                    if is_rate_limited(error_msg) or is_api_key_error(error_msg):
+                        self.skipTest(f"Anthropic streaming error: {error_msg}")
+                    raise
+
+        asyncio.run(_async_test())
+
+    def test_react_tool_execution_error_handling(self):
+        """Test ReAct agent handling of tool execution errors."""
+
+        def error_tool(x: float) -> float:
+            """A tool that intentionally raises an error."""
+            if x < 0:
+                raise ValueError("Cannot process negative numbers")
+            return x * 2
+
+        error_tools = [ToolsFactory().create_tool(error_tool)]
+
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=error_tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            # Test with valid input first
+            response1 = agent.chat("Use the error_tool with input 5.")
+            self.check_response_and_skip(response1, "Anthropic")
+
+            if not is_rate_limited(response1.response):
+                self.assertIn("10", response1.response)  # 5 * 2 = 10
+
+            # Test with invalid input that should cause tool error
+            response2 = agent.chat("Use the error_tool with input -1.")
+            self.check_response_and_skip(response2, "Anthropic")
+
+            if not is_rate_limited(response2.response):
+                # ReAct agent should handle the tool error gracefully
+                # and provide some kind of error message or explanation
+                self.assertTrue(len(response2.response) > 0)
+                # Should mention error or inability to process
+                error_indicators = ["error", "cannot", "unable", "negative", "problem"]
+                has_error_indication = any(
+                    indicator in response2.response.lower()
+                    for indicator in error_indicators
+                )
+                self.assertTrue(
+                    has_error_indication,
+                    f"Response should indicate error handling: {response2.response}",
+                )
+
+    def test_react_workflow_interruption_handling(self):
+        """Test ReAct agent handling of workflow interruptions."""
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            # Test a complex multi-step task that might be interrupted
+            response = agent.chat(
+                "Calculate 3 times 7, then multiply that by 4, then add 10 to the result."
+            )
+            self.check_response_and_skip(response, "Anthropic")
+
+            if response.response and not is_rate_limited(response.response):
+                # Final result should be: (3*7)*4+10 = 21*4+10 = 84+10 = 94
+                self.assertIn("94", response.response)
+
+    def test_react_memory_corruption_recovery(self):
+        """Test ReAct agent recovery from memory-related errors."""
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            # Normal operation
+            response1 = agent.chat("Calculate 6 times 7.")
+            self.check_response_and_skip(response1, "Anthropic")
+
+            if not is_rate_limited(response1.response):
+                self.assertIn("42", response1.response)
+
+                # Continue conversation to test memory consistency
+                response2 = agent.chat(
+                    "What was the result of the previous calculation?"
+                )
+                self.check_response_and_skip(response2, "Anthropic")
+
+                if not is_rate_limited(response2.response):
+                    # Should remember the previous result
+                    self.assertIn("42", response2.response)
+
+    def test_react_fallback_behavior_on_provider_failure(self):
+        """Test ReAct agent behavior when provider fails completely."""
+
+        # Create a config with an invalid API configuration to simulate failure
+        invalid_config = AgentConfig(
+            agent_type=AgentType.REACT,
+            main_llm_provider=ModelProvider.ANTHROPIC,
+            tool_llm_provider=ModelProvider.ANTHROPIC,
+        )
+
+        agent = Agent(
+            agent_config=invalid_config,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            try:
+                response = agent.chat("What is 2 times 3?")
+                self.check_response_and_skip(response, "Anthropic")
+
+                # If we get here without error, the test passed
+                if response.response and not is_rate_limited(response.response):
+                    self.assertIn("6", response.response)
+
+            except Exception as e:
+                error_msg = str(e)
+                if is_rate_limited(error_msg) or is_api_key_error(error_msg):
+                    self.skipTest(f"Anthropic provider failure: {error_msg}")
+                # For other exceptions, let them bubble up as actual test failures
+                raise
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_react_memory.py

@@ -0,0 +1,257 @@
+# Suppress external dependency warnings before any other imports
+import warnings
+
+warnings.simplefilter("ignore", DeprecationWarning)
+
+import unittest
+import threading
+
+from vectara_agentic.agent import Agent
+from vectara_agentic.tools import ToolsFactory
+from llama_index.core.llms import MessageRole
+
+import nest_asyncio
+
+nest_asyncio.apply()
+
+from conftest import (
+    AgentTestMixin,
+    react_config_anthropic,
+    react_config_gemini,
+    mult,
+    add,
+    STANDARD_TEST_TOPIC,
+    STANDARD_TEST_INSTRUCTIONS,
+)
+
+ARIZE_LOCK = threading.Lock()
+
+
+class TestReActMemory(unittest.TestCase, AgentTestMixin):
+    """Test memory persistence and conversation history for ReAct agents."""
+
+    def setUp(self):
+        self.tools = [ToolsFactory().create_tool(mult), ToolsFactory().create_tool(add)]
+        self.topic = STANDARD_TEST_TOPIC
+        self.instructions = STANDARD_TEST_INSTRUCTIONS
+        self.session_id = "test-react-memory-123"
+
+    def test_react_memory_persistence_across_chats(self):
+        """Test that ReAct agents maintain conversation context across multiple chat calls."""
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+            session_id=self.session_id,
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            # First interaction - establish context
+            response1 = agent.chat("Calculate 5 times 10 and remember this result.")
+            self.check_response_and_skip(response1, "Anthropic")
+            self.assertIn("50", response1.response)
+
+            # Second interaction - reference previous result
+            response2 = agent.chat(
+                "What was the result I asked you to calculate and remember in the previous message?"
+            )
+            self.check_response_and_skip(response2, "Anthropic")
+            self.assertIn("50", response2.response)
+
+            # Third interaction - use previous result in new calculation
+            response3 = agent.chat("Add 25 to the number you calculated earlier.")
+            self.check_response_and_skip(response3, "Anthropic")
+            # Should be 50 + 25 = 75
+            self.assertIn("75", response3.response)
+
+    def test_react_memory_with_tool_history(self):
+        """Test that ReAct agents remember tool usage history across conversations."""
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+            session_id=self.session_id + "_tools",
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            # Use multiplication tool
+            response1 = agent.chat("Multiply 7 by 8.")
+            self.check_response_and_skip(response1, "Anthropic")
+            self.assertIn("56", response1.response)
+
+            # Use addition tool
+            response2 = agent.chat("Add 20 to 30.")
+            self.check_response_and_skip(response2, "Anthropic")
+            self.assertIn("50", response2.response)
+
+            # Reference both previous tool uses
+            response3 = agent.chat(
+                "What were the two calculations I asked you to perform? "
+                "Add those two results together."
+            )
+            self.check_response_and_skip(response3, "Anthropic")
+            # Should remember 7*8=56 and 20+30=50, then 56+50=106
+            self.assertIn("106", response3.response)
+
+    def test_react_memory_state_consistency(self):
+        """Test that ReAct agent memory state remains consistent during workflow execution."""
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+            session_id=self.session_id + "_consistency",
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            # Check initial memory state
+            initial_memory_size = len(agent.memory.get_all())
+
+            # Perform a conversation with tool use
+            response1 = agent.chat("Calculate 6 times 9 and tell me the result.")
+            self.check_response_and_skip(response1, "Anthropic")
+
+            # Memory should now contain user message, tool calls, and assistant response
+            after_first_memory = agent.memory.get_all()
+            self.assertGreater(
+                len(after_first_memory),
+                initial_memory_size,
+                "Memory should contain new messages after interaction",
+            )
+
+            # Continue conversation
+            response2 = agent.chat("Double that result.")
+            self.check_response_and_skip(response2, "Anthropic")
+
+            # Memory should contain all previous messages plus new ones
+            after_second_memory = agent.memory.get_all()
+            self.assertGreater(
+                len(after_second_memory),
+                len(after_first_memory),
+                "Memory should accumulate messages across interactions",
+            )
+
+            # Final result should be (6*9)*2 = 54*2 = 108
+            self.assertIn("108", response2.response)
+
+    def test_react_memory_across_different_providers(self):
+        """Test memory consistency when using different ReAct providers."""
+        # Test with Anthropic
+        agent_anthropic = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+            session_id=self.session_id + "_anthropic_provider",
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            response1 = agent_anthropic.chat("Multiply 4 by 12.")
+            self.check_response_and_skip(response1, "Anthropic")
+            self.assertIn("48", response1.response)
+
+            # Verify memory structure is consistent
+            anthropic_memory = agent_anthropic.memory.get_all()
+            self.assertGreater(len(anthropic_memory), 0)
+
+        # Test with Gemini (if available)
+        agent_gemini = Agent(
+            agent_config=react_config_gemini,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+            session_id=self.session_id + "_gemini_provider",
+        )
+
+        with self.with_provider_fallback("Gemini"):
+            response2 = agent_gemini.chat("Multiply 4 by 12.")
+            self.check_response_and_skip(response2, "Gemini")
+            self.assertIn("48", response2.response)
+
+            # Verify memory structure is consistent across providers
+            gemini_memory = agent_gemini.memory.get_all()
+            self.assertGreater(len(gemini_memory), 0)
+
+    def test_react_memory_serialization_compatibility(self):
+        """Test that ReAct agent memory can be properly serialized and deserialized."""
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+            session_id=self.session_id + "_serialization",
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            # Perform some interactions to populate memory
+            response1 = agent.chat("Calculate 15 times 3.")
+            self.check_response_and_skip(response1, "Anthropic")
+
+            response2 = agent.chat("Add 10 to the previous result.")
+            self.check_response_and_skip(response2, "Anthropic")
+
+            # Get memory state before serialization
+            original_memory = agent.memory.get_all()
+            original_memory_size = len(original_memory)
+
+            # Test that memory state can be accessed and contains expected content
+            self.assertGreater(original_memory_size, 0)
+
+            # Verify memory contains both user and assistant messages
+            user_messages = [
+                msg for msg in original_memory if msg.role == MessageRole.USER
+            ]
+            assistant_messages = [
+                msg for msg in original_memory if msg.role == MessageRole.ASSISTANT
+            ]
+
+            self.assertGreater(
+                len(user_messages), 0, "Should have user messages in memory"
+            )
+            self.assertGreater(
+                len(assistant_messages), 0, "Should have assistant messages in memory"
+            )
+
+    async def test_react_memory_async_streaming_consistency(self):
+        """Test memory consistency during async streaming operations with ReAct agents."""
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+            session_id=self.session_id + "_async_streaming",
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            # First streaming interaction
+            stream1 = await agent.astream_chat("Calculate 8 times 7.")
+            async for chunk in stream1.async_response_gen():
+                pass
+            response1 = await stream1.aget_response()
+            self.check_response_and_skip(response1, "Anthropic")
+
+            # Check memory after first streaming interaction
+            memory_after_first = agent.memory.get_all()
+            self.assertGreater(len(memory_after_first), 0)
+
+            # Second streaming interaction that references the first
+            stream2 = await agent.astream_chat(
+                "Subtract 6 from the result you just calculated."
+            )
+            async for chunk in stream2.async_response_gen():
+                pass
+            response2 = await stream2.aget_response()
+            self.check_response_and_skip(response2, "Anthropic")
+
+            # Memory should contain both interactions
+            memory_after_second = agent.memory.get_all()
+            self.assertGreater(len(memory_after_second), len(memory_after_first))
+
+            # Final result should be (8*7)-6 = 56-6 = 50
+            self.assertIn("50", response2.response)
+
+
+if __name__ == "__main__":
+    unittest.main()

tests/test_react_streaming.py

@@ -0,0 +1,135 @@
+# Suppress external dependency warnings before any other imports
+import warnings
+
+warnings.simplefilter("ignore", DeprecationWarning)
+
+import unittest
+
+from vectara_agentic.agent import Agent
+from vectara_agentic.tools import ToolsFactory
+
+import nest_asyncio
+
+nest_asyncio.apply()
+
+from conftest import (
+    AgentTestMixin,
+    react_config_openai,
+    react_config_anthropic,
+    react_config_gemini,
+    react_config_together,
+    mult,
+    STANDARD_TEST_TOPIC,
+    STANDARD_TEST_INSTRUCTIONS,
+)
+
+
+class TestReActStreaming(unittest.IsolatedAsyncioTestCase, AgentTestMixin):
+    """Test streaming functionality for ReAct agents across all providers."""
+
+    def setUp(self):
+        self.tools = [ToolsFactory().create_tool(mult)]
+        self.topic = STANDARD_TEST_TOPIC
+        self.instructions = STANDARD_TEST_INSTRUCTIONS
+
+    async def _test_react_streaming_workflow(self, config, provider_name):
+        """Common workflow for testing ReAct streaming with any provider."""
+        agent = Agent(
+            agent_config=config,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions=self.instructions,
+        )
+
+        with self.with_provider_fallback(provider_name):
+            # First calculation: 5 * 10 = 50
+            stream1 = await agent.astream_chat(
+                "What is 5 times 10. Only give the answer, nothing else"
+            )
+            # Consume the stream
+            async for chunk in stream1.async_response_gen():
+                pass
+            response1 = await stream1.aget_response()
+            self.check_response_and_skip(response1, provider_name)
+
+            # Second calculation: 3 * 7 = 21
+            stream2 = await agent.astream_chat(
+                "what is 3 times 7. Only give the answer, nothing else"
+            )
+            # Consume the stream
+            async for chunk in stream2.async_response_gen():
+                pass
+            response2 = await stream2.aget_response()
+            self.check_response_and_skip(response2, provider_name)
+
+            # Final calculation: 50 * 21 = 1050
+            stream3 = await agent.astream_chat(
+                "multiply the results of the last two multiplications. Only give the answer, nothing else."
+            )
+            # Consume the stream and collect chunks for verification
+            chunks = []
+            async for chunk in stream3.async_response_gen():
+                chunks.append(chunk)
+
+            response3 = await stream3.aget_response()
+            self.check_response_and_skip(response3, provider_name)
+
+            # Verify the final result
+            self.assertIn("1050", response3.response)
+
+            # Verify we actually got streaming chunks
+            self.assertGreater(
+                len(chunks), 0, f"{provider_name} should produce streaming chunks"
+            )
+
+    async def test_anthropic_react_streaming(self):
+        """Test ReAct agent streaming with Anthropic."""
+        await self._test_react_streaming_workflow(react_config_anthropic, "Anthropic")
+
+    async def test_openai_react_streaming(self):
+        """Test ReAct agent streaming with OpenAI."""
+        await self._test_react_streaming_workflow(react_config_openai, "OpenAI")
+
+    async def test_gemini_react_streaming(self):
+        """Test ReAct agent streaming with Gemini."""
+        await self._test_react_streaming_workflow(react_config_gemini, "Gemini")
+
+    async def test_together_react_streaming(self):
+        """Test ReAct agent streaming with Together.AI."""
+        await self._test_react_streaming_workflow(react_config_together, "Together AI")
+
+    async def test_react_streaming_reasoning_pattern(self):
+        """Test that ReAct agents demonstrate reasoning patterns in streaming responses."""
+        agent = Agent(
+            agent_config=react_config_anthropic,
+            tools=self.tools,
+            topic=self.topic,
+            custom_instructions="Think step by step and show your reasoning before using tools.",
+        )
+
+        with self.with_provider_fallback("Anthropic"):
+            # Ask a question that requires multi-step reasoning
+            stream = await agent.astream_chat(
+                "I need to calculate 7 times 8, then add 12 to that result, then multiply by 2. "
+                "Show me your reasoning process."
+            )
+
+            chunks = []
+            async for chunk in stream.async_response_gen():
+                chunks.append(str(chunk))
+
+            response = await stream.aget_response()
+            self.check_response_and_skip(response, "Anthropic")
+
+            # Verify we got streaming content
+            self.assertGreater(len(chunks), 0)
+
+            # For ReAct agents, we should see reasoning patterns in the response
+            full_content = "".join(chunks).lower()
+
+            # The final answer should be correct: (7*8 + 12) * 2 = (56 + 12) * 2 = 68 * 2 = 136
+            self.assertTrue("136" in response.response or "136" in full_content)
+
+
+if __name__ == "__main__":
+    unittest.main()