vectara-agentic 0.4.4__tar.gz → 0.4.10__tar.gz

{vectara_agentic-0.4.4/vectara_agentic.egg-info → vectara_agentic-0.4.10}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.4
 Name: vectara_agentic
-Version: 0.4.4
+Version: 0.4.10
 Summary: A Python package for creating AI Assistants and AI Agents with Vectara
 Home-page: https://github.com/vectara/py-vectara-agentic
 Author: Ofer Mendelevitch
@@ -16,41 +16,43 @@ Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: llama-index==0.13.3
-Requires-Dist: llama-index-core==0.13.3
-Requires-Dist: llama-index-workflows==1.3.0
-Requires-Dist: llama-index-cli==0.5.0
-Requires-Dist: llama-index-indices-managed-vectara==0.5.0
-Requires-Dist: llama-index-llms-openai==0.5.4
-Requires-Dist: llama-index-llms-openai-like==0.5.0
-Requires-Dist: llama-index-llms-anthropic==0.8.5
-Requires-Dist: llama-index-llms-together==0.4.0
-Requires-Dist: llama-index-llms-groq==0.4.0
-Requires-Dist: llama-index-llms-cohere==0.6.0
-Requires-Dist: llama-index-llms-google-genai==0.3.0
-Requires-Dist: google_genai>=1.31.0
-Requires-Dist: llama-index-llms-bedrock-converse==0.8.2
-Requires-Dist: llama-index-tools-yahoo-finance==0.4.0
-Requires-Dist: llama-index-tools-arxiv==0.4.0
-Requires-Dist: llama-index-tools-database==0.4.0
-Requires-Dist: llama-index-tools-google==0.6.0
-Requires-Dist: llama-index-tools-tavily_research==0.4.0
-Requires-Dist: llama_index.tools.brave_search==0.4.0
-Requires-Dist: llama-index-tools-neo4j==0.4.0
-Requires-Dist: llama-index-tools-waii==0.4.0
-Requires-Dist: llama-index-graph-stores-kuzu==0.9.0
-Requires-Dist: llama-index-tools-salesforce==0.4.0
-Requires-Dist: llama-index-tools-slack==0.4.0
-Requires-Dist: llama-index-tools-exa==0.4.0
-Requires-Dist: llama-index-tools-wikipedia==0.4.0
-Requires-Dist: llama-index-tools-bing-search==0.4.0
+Requires-Dist: llama-index==0.14.7
+Requires-Dist: llama-index-core==0.14.7
+Requires-Dist: llama-index-workflows==2.10.3
+Requires-Dist: llama-index-cli==0.5.3
+Requires-Dist: llama-index-indices-managed-vectara==0.5.1
+Requires-Dist: llama-index-llms-openai==0.6.7
+Requires-Dist: llama-index-llms-openai-like==0.5.3
+Requires-Dist: llama-index-llms-anthropic==0.9.7
+Requires-Dist: llama-index-llms-together==0.4.1
+Requires-Dist: llama-index-llms-groq==0.4.1
+Requires-Dist: llama-index-llms-cohere==0.6.1
+Requires-Dist: llama-index-llms-google-genai==0.7.1
+Requires-Dist: google_genai==1.48.0
+Requires-Dist: llama-index-llms-bedrock-converse==0.11.0
+Requires-Dist: llama-index-tools-yahoo-finance==0.4.1
+Requires-Dist: llama-index-tools-arxiv==0.4.1
+Requires-Dist: llama-index-tools-database==0.4.1
+Requires-Dist: llama-index-tools-google==0.6.2
+Requires-Dist: llama-index-tools-tavily_research==0.4.1
+Requires-Dist: llama_index.tools.brave_search==0.4.1
+Requires-Dist: llama-index-tools-neo4j==0.4.1
+Requires-Dist: llama-index-tools-waii==0.4.1
+Requires-Dist: llama-index-graph-stores-kuzu==0.9.1
+Requires-Dist: llama-index-tools-salesforce==0.4.1
+Requires-Dist: llama-index-tools-slack==0.4.1
+Requires-Dist: llama-index-tools-exa==0.4.1
+Requires-Dist: llama-index-tools-wikipedia==0.4.1
+Requires-Dist: llama-index-tools-bing-search==0.4.1
 Requires-Dist: openai>=1.99.3
 Requires-Dist: tavily-python>=0.7.10
 Requires-Dist: exa-py>=1.14.20
-Requires-Dist: openinference-instrumentation-llama-index==4.3.4
+Requires-Dist: openinference-instrumentation-llama-index==4.3.8
 Requires-Dist: opentelemetry-proto>=1.31.0
-Requires-Dist: arize-phoenix==10.9.1
-Requires-Dist: arize-phoenix-otel==0.10.3
+Requires-Dist: arize-phoenix==12.9.0
+Requires-Dist: arize-phoenix-otel==0.13.1
+Requires-Dist: arize-phoenix-client==1.21.0
+Requires-Dist: arize-phoenix-evals==2.5.0
 Requires-Dist: protobuf==5.29.5
 Requires-Dist: tokenizers>=0.20
 Requires-Dist: pydantic>=2.11.5
@@ -736,13 +738,13 @@ If you want to use `agent`, `tools`, `llm` or `verbose` in other events (that ar
 the `Context` of the Workflow as follows:
 
 ```python
-await ctx.set("agent", ev.agent)
+await ctx.store.set("agent", ev.agent)
 ```
 
 and then in any other event you can pull that agent object with
 
 ```python
-agent = await ctx.get("agent")
+agent = await ctx.store.get("agent")
 ```
 
 Similarly you can reuse the `llm`, `tools` or `verbose` arguments within other nodes in the workflow.
@@ -868,7 +870,7 @@ from vectara_agentic import AgentConfig, AgentType, ModelProvider
 agent_config = AgentConfig(
     agent_type = AgentType.REACT,
     main_llm_provider = ModelProvider.ANTHROPIC,
-    main_llm_model_name = 'claude-3-5-sonnet-20241022',
+    main_llm_model_name = 'claude-4-5-sonnet',
     tool_llm_provider = ModelProvider.TOGETHER,
     tool_llm_model_name = 'deepseek-ai/DeepSeek-V3'
 )
@@ -886,7 +888,7 @@ The `AgentConfig` object may include the following items:
 - `main_llm_provider` and `tool_llm_provider`: the LLM provider for main agent and for the tools. Valid values are `OPENAI`, `ANTHROPIC`, `TOGETHER`, `GROQ`, `COHERE`, `BEDROCK`, `GEMINI` (default: `OPENAI`).
 
 > **Note:** Fireworks AI support has been removed. If you were using Fireworks, please migrate to one of the supported providers listed above.
-- `main_llm_model_name` and `tool_llm_model_name`: agent model name for agent and tools (default depends on provider: OpenAI uses gpt-4.1-mini, Gemini uses gemini-2.5-flash-lite).
+- `main_llm_model_name` and `tool_llm_model_name`: agent model name for agent and tools (default depends on provider: OpenAI uses gpt-4.1-mini, Anthropic uses claude-sonnet-4-5, Gemini uses models/gemini-2.5-flash, Together.AI uses deepseek-ai/DeepSeek-V3, GROQ uses openai/gpt-oss-20b, Bedrock uses us.anthropic.claude-sonnet-4-20250514-v1:0, Cohere uses command-a-03-2025).
 - `observer`: the observer type; should be `ARIZE_PHOENIX` or if undefined no observation framework will be used.
 - `endpoint_api_key`: a secret key if using the API endpoint option (defaults to `dev-api-key`)
 
@@ -913,6 +915,7 @@ config = AgentConfig(
     main_llm_model_name="meta-llama/Meta-Llama-3.1-8B-Instruct",
     private_llm_api_base="http://vllm-server.company.com/v1",
     private_llm_api_key="TEST_API_KEY",
+    private_llm_max_tokens=8192,  # Optional: set max output tokens for your private LLM
 )
 
 agent = Agent(

{vectara_agentic-0.4.4 → vectara_agentic-0.4.10}/README.md

@@ -661,13 +661,13 @@ If you want to use `agent`, `tools`, `llm` or `verbose` in other events (that ar
 the `Context` of the Workflow as follows:
 
 ```python
-await ctx.set("agent", ev.agent)
+await ctx.store.set("agent", ev.agent)
 ```
 
 and then in any other event you can pull that agent object with
 
 ```python
-agent = await ctx.get("agent")
+agent = await ctx.store.get("agent")
 ```
 
 Similarly you can reuse the `llm`, `tools` or `verbose` arguments within other nodes in the workflow.
@@ -793,7 +793,7 @@ from vectara_agentic import AgentConfig, AgentType, ModelProvider
 agent_config = AgentConfig(
     agent_type = AgentType.REACT,
     main_llm_provider = ModelProvider.ANTHROPIC,
-    main_llm_model_name = 'claude-3-5-sonnet-20241022',
+    main_llm_model_name = 'claude-4-5-sonnet',
     tool_llm_provider = ModelProvider.TOGETHER,
     tool_llm_model_name = 'deepseek-ai/DeepSeek-V3'
 )
@@ -811,7 +811,7 @@ The `AgentConfig` object may include the following items:
 - `main_llm_provider` and `tool_llm_provider`: the LLM provider for main agent and for the tools. Valid values are `OPENAI`, `ANTHROPIC`, `TOGETHER`, `GROQ`, `COHERE`, `BEDROCK`, `GEMINI` (default: `OPENAI`).
 
 > **Note:** Fireworks AI support has been removed. If you were using Fireworks, please migrate to one of the supported providers listed above.
-- `main_llm_model_name` and `tool_llm_model_name`: agent model name for agent and tools (default depends on provider: OpenAI uses gpt-4.1-mini, Gemini uses gemini-2.5-flash-lite).
+- `main_llm_model_name` and `tool_llm_model_name`: agent model name for agent and tools (default depends on provider: OpenAI uses gpt-4.1-mini, Anthropic uses claude-sonnet-4-5, Gemini uses models/gemini-2.5-flash, Together.AI uses deepseek-ai/DeepSeek-V3, GROQ uses openai/gpt-oss-20b, Bedrock uses us.anthropic.claude-sonnet-4-20250514-v1:0, Cohere uses command-a-03-2025).
 - `observer`: the observer type; should be `ARIZE_PHOENIX` or if undefined no observation framework will be used.
 - `endpoint_api_key`: a secret key if using the API endpoint option (defaults to `dev-api-key`)
 
@@ -838,6 +838,7 @@ config = AgentConfig(
     main_llm_model_name="meta-llama/Meta-Llama-3.1-8B-Instruct",
     private_llm_api_base="http://vllm-server.company.com/v1",
     private_llm_api_key="TEST_API_KEY",
+    private_llm_max_tokens=8192,  # Optional: set max output tokens for your private LLM
 )
 
 agent = Agent(

vectara_agentic-0.4.10/requirements.txt

@@ -0,0 +1,46 @@
+llama-index==0.14.7
+llama-index-core==0.14.7
+llama-index-workflows==2.10.3
+llama-index-cli==0.5.3
+llama-index-indices-managed-vectara==0.5.1
+llama-index-llms-openai==0.6.7
+llama-index-llms-openai-like==0.5.3
+llama-index-llms-anthropic==0.9.7
+llama-index-llms-together==0.4.1
+llama-index-llms-groq==0.4.1
+llama-index-llms-cohere==0.6.1
+llama-index-llms-google-genai==0.7.1
+google_genai==1.48.0
+llama-index-llms-bedrock-converse==0.11.0
+llama-index-tools-yahoo-finance==0.4.1
+llama-index-tools-arxiv==0.4.1
+llama-index-tools-database==0.4.1
+llama-index-tools-google==0.6.2
+llama-index-tools-tavily_research==0.4.1
+llama_index.tools.brave_search==0.4.1
+llama-index-tools-neo4j==0.4.1
+llama-index-tools-waii==0.4.1
+llama-index-graph-stores-kuzu==0.9.1
+llama-index-tools-salesforce==0.4.1
+llama-index-tools-slack==0.4.1
+llama-index-tools-exa==0.4.1
+llama-index-tools-wikipedia==0.4.1
+llama-index-tools-bing-search==0.4.1
+openai>=1.99.3
+tavily-python>=0.7.10
+exa-py>=1.14.20
+openinference-instrumentation-llama-index==4.3.8
+opentelemetry-proto>=1.31.0
+arize-phoenix==12.9.0
+arize-phoenix-otel==0.13.1
+arize-phoenix-client==1.21.0
+arize-phoenix-evals==2.5.0
+protobuf==5.29.5
+tokenizers>=0.20
+pydantic>=2.11.5
+pandas==2.2.3
+retrying==1.4.2
+python-dotenv==1.0.1
+cloudpickle>=3.1.1
+httpx==0.28.1
+commonmark==0.9.1

{vectara_agentic-0.4.4 → vectara_agentic-0.4.10}/tests/benchmark_models.py

@@ -68,7 +68,7 @@ def validate_api_keys(models_to_test: List[Dict]) -> None:
             missing_keys.append(key)
 
     if missing_keys:
-        print("❌ ERROR: Missing required API keys for benchmark execution:")
+        print("ERROR: Missing required API keys for benchmark execution:")
         print()
         for key in sorted(missing_keys):
             print(f"  • {key}")
@@ -83,7 +83,7 @@ def validate_api_keys(models_to_test: List[Dict]) -> None:
 
         sys.exit(1)
 
-    print("✅ All required API keys are present")
+    print("All required API keys are present")
     print(f"Found API keys for {len(required_keys)} required environment variables")
 
 
@@ -135,7 +135,7 @@ class ModelBenchmark:
             {"provider": ModelProvider.OPENAI, "model": "gpt-5-mini"},
             {"provider": ModelProvider.OPENAI, "model": "gpt-4o-mini"},
             {"provider": ModelProvider.OPENAI, "model": "gpt-4.1-mini"},
-            {"provider": ModelProvider.ANTHROPIC, "model": "claude-sonnet-4-20250514"},
+            {"provider": ModelProvider.ANTHROPIC, "model": "claude-sonnet-4-5"},
             {"provider": ModelProvider.TOGETHER, "model": "deepseek-ai/DeepSeek-V3"},
             {"provider": ModelProvider.GROQ, "model": "openai/gpt-oss-20b"},
             {"provider": ModelProvider.GEMINI, "model": "models/gemini-2.5-flash-lite"},
@@ -817,11 +817,11 @@ class ModelBenchmark:
             observability_setup = setup_observer(dummy_config, verbose=True)
             if observability_setup:
                 print(
-                    "✅ Arize Phoenix observability enabled - LLM calls will be traced\n"
+                    "Arize Phoenix observability enabled - LLM calls will be traced\n"
                 )
                 _observability_initialized = True
             else:
-                print("⚠️  Arize Phoenix observability setup failed\n")
+                print("Arize Phoenix observability setup failed\n")
 
         # Create semaphore to limit concurrent model testing
         model_semaphore = asyncio.Semaphore(self.max_concurrent_models)
@@ -835,7 +835,7 @@ class ModelBenchmark:
             tasks.append(task)
 
         # Execute all model benchmarks in parallel
-        print("🚀 Starting parallel benchmark execution...\n")
+        print("Starting parallel benchmark execution...\n")
         await asyncio.gather(*tasks, return_exceptions=True)
 
     async def _run_model_benchmark(
@@ -857,9 +857,9 @@ class ModelBenchmark:
                         provider, model_name, test_name, test_config
                     )
                 except Exception as e:
-                    print(f"❌ Error in {model_name} - {test_name}: {e}")
+                    print(f"Error in {model_name} - {test_name}: {e}")
 
-            print(f"✅ Completed: {provider.value} - {model_name}")
+            print(f"Completed: {provider.value} - {model_name}")
 
     async def _run_scenario_benchmark(
         self,
@@ -892,18 +892,18 @@ class ModelBenchmark:
 
                 if result.error:
                     print(
-                        f"    ❌ {model_name}/{test_name} Iteration {iteration_num}: {result.error}"
+                        f"{model_name}/{test_name} Iteration {iteration_num}: {result.error}"
                     )
                 else:
                     print(
-                        f"    ✅ {model_name}/{test_name} Iteration {iteration_num}: "
+                        f"{model_name}/{test_name} Iteration {iteration_num}: "
                         f"{result.total_response_time:.2f}s, "
                         f"first token: {result.first_token_latency:.2f}s, "
                         f"{result.tokens_per_second:.1f} chars/sec"
                     )
 
             except Exception as e:
-                print(f"    ❌ {model_name}/{test_name} Iteration {iteration_num}: {e}")
+                print(f"{model_name}/{test_name} Iteration {iteration_num}: {e}")
                 # Create error result
                 error_result = BenchmarkResult(
                     model_name=model_name,
@@ -929,7 +929,7 @@ class ModelBenchmark:
         successful = len([r for r in iteration_results if r.error is None])
         success_rate = (successful / len(iteration_results)) * 100
         print(
-            f"    📊 {model_name}/{test_name} complete: {successful}/{len(iteration_results)} successful ({success_rate:.1f}%)"
+            f"{model_name}/{test_name} complete: {successful}/{len(iteration_results)} successful ({success_rate:.1f}%)"
         )
 
         return iteration_results

{vectara_agentic-0.4.4 → vectara_agentic-0.4.10}/tests/endpoint.py

@@ -1,6 +1,7 @@
 #!/usr/bin/env python3
 import json
 import logging
+import os
 from functools import wraps
 from flask import Flask, request, Response, jsonify
 from openai import OpenAI
@@ -16,6 +17,15 @@ werkzeug_log.setLevel(logging.ERROR)
 # Load expected API key from environment (fallback for testing)
 EXPECTED_API_KEY = "TEST_API_KEY"
 
+# Together.AI models that should be routed to Together.AI
+TOGETHER_MODELS = [
+    "openai/gpt-oss-120b",
+    "openai/gpt-oss-20b",
+    "deepseek-ai/DeepSeek-V3",
+    "deepseek-ai/DeepSeek-V3.1",
+    "deepseek-ai/DeepSeek-R1",
+]
+
 
 # Authentication decorator
 def require_api_key(f):
@@ -41,7 +51,8 @@ def log_request_info():
 @require_api_key
 def chat_completions():
     """
-    Proxy endpoint for OpenAI Chat Completions.
+    Proxy endpoint for OpenAI-compatible Chat Completions.
+    Routes to different backends based on model name.
     Supports both streaming and non-streaming modes.
     """
     try:
@@ -49,7 +60,21 @@ def chat_completions():
     except Exception:
         return jsonify({"error": "Invalid JSON payload"}), 400
 
-    client = OpenAI()
+    # Determine which client to use based on model name
+    model_name = data.get("model", "")
+    if model_name in TOGETHER_MODELS:
+        # Route to Together.AI
+        together_api_key = os.getenv("TOGETHER_API_KEY")
+        if not together_api_key:
+            return jsonify({"error": "TOGETHER_API_KEY environment variable not set"}), 500
+        client = OpenAI(
+            api_key=together_api_key,
+            base_url="https://api.together.xyz/v1"
+        )
+    else:
+        # Default to OpenAI
+        client = OpenAI()
+
     is_stream = data.get("stream", False)
 
     if is_stream:

{vectara_agentic-0.4.4 → vectara_agentic-0.4.10}/tests/test_agent.py

@@ -13,7 +13,6 @@ from vectara_agentic.agent_config import AgentConfig
 from vectara_agentic.types import ModelProvider, ObserverType
 from vectara_agentic.tools import ToolsFactory
 
-from vectara_agentic.agent_core.prompts import GENERAL_INSTRUCTIONS
 from conftest import mult, STANDARD_TEST_TOPIC, STANDARD_TEST_INSTRUCTIONS
 
 
@@ -54,9 +53,11 @@ class TestAgentPackage(unittest.TestCase):
             + date.today().strftime("%A, %B %d, %Y")
             + " with Always do as your mother tells you!"
         )
+        # Test format_prompt with dummy instructions since we're only testing template substitution
+        dummy_instructions = "Test instructions"
         self.assertEqual(
             format_prompt(
-                prompt_template, GENERAL_INSTRUCTIONS, topic, custom_instructions
+                prompt_template, dummy_instructions, topic, custom_instructions
             ),
             expected_output,
         )
@@ -83,7 +84,7 @@ class TestAgentPackage(unittest.TestCase):
             config = AgentConfig(
                 agent_type=AgentType.REACT,
                 main_llm_provider=ModelProvider.ANTHROPIC,
-                main_llm_model_name="claude-sonnet-4-20250514",
+                main_llm_model_name="claude-sonnet-4-5",
                 tool_llm_provider=ModelProvider.TOGETHER,
                 tool_llm_model_name="moonshotai/Kimi-K2-Instruct",
                 observer=ObserverType.ARIZE_PHOENIX,

vectara_agentic-0.4.10/tests/test_bedrock.py

@@ -0,0 +1,170 @@
+# 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 vectara_agentic.tools_catalog import ToolsCatalog
+
+import nest_asyncio
+
+nest_asyncio.apply()
+
+from conftest import (
+    mult,
+    add,
+    fc_config_bedrock,
+    STANDARD_TEST_TOPIC,
+    STANDARD_TEST_INSTRUCTIONS,
+)
+
+ARIZE_LOCK = threading.Lock()
+
+
+class TestBedrock(unittest.IsolatedAsyncioTestCase):
+
+    async def test_multiturn(self):
+        with ARIZE_LOCK:
+            tools = [ToolsFactory().create_tool(mult)]
+            agent = Agent(
+                tools=tools,
+                topic=STANDARD_TEST_TOPIC,
+                custom_instructions=STANDARD_TEST_INSTRUCTIONS,
+                agent_config=fc_config_bedrock,
+            )
+
+            # 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
+            _ = await stream1.aget_response()
+
+            # 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
+            _ = await stream2.aget_response()
+
+            # Final calculation: 50 * 21 = 1050
+            stream3 = await agent.astream_chat(
+                "multiply the results of the last two questions. Output only the answer."
+            )
+            # Consume the stream
+            async for chunk in stream3.async_response_gen():
+                pass
+            response3 = await stream3.aget_response()
+
+            self.assertEqual(response3.response, "1050")
+
+    async def test_claude_sonnet_4_multi_tool_chain(self):
+        """Test Claude Sonnet 4 with complex multi-step reasoning chain using multiple tools via Bedrock."""
+        with ARIZE_LOCK:
+            # Use Bedrock config (Claude Sonnet 4)
+            tools_catalog = ToolsCatalog(fc_config_bedrock)
+            tools = [
+                ToolsFactory().create_tool(mult),
+                ToolsFactory().create_tool(add),
+                ToolsFactory().create_tool(tools_catalog.summarize_text),
+                ToolsFactory().create_tool(tools_catalog.rephrase_text),
+            ]
+
+            agent = Agent(
+                agent_config=fc_config_bedrock,
+                tools=tools,
+                topic=STANDARD_TEST_TOPIC,
+                custom_instructions="You are a mathematical reasoning agent that explains your work step by step.",
+            )
+
+            # Complex multi-step reasoning task
+            complex_query = (
+                "Perform this calculation step by step: "
+                "First multiply 5 by 9, then add 13 to that result, "
+                "then multiply the new result by 2. "
+                "After getting the final number, summarize the entire mathematical process "
+                "with expertise in 'mathematics education', "
+                "then rephrase that summary as a 10-year-old would explain it."
+            )
+
+            print("\nStarting Claude Sonnet 4 multi-tool chain test (Bedrock)")
+            print(f"Query: {complex_query}")
+            print("Streaming response:\n" + "="*50)
+
+            stream = await agent.astream_chat(complex_query)
+
+            # Capture streaming deltas and tool calls
+            streaming_deltas = []
+            tool_calls_made = []
+            full_response = ""
+
+            async for chunk in stream.async_response_gen():
+                if chunk and chunk.strip():
+                    streaming_deltas.append(chunk)
+                    full_response += chunk
+                    # Display each streaming delta
+                    print(f"Delta: {repr(chunk)}")
+
+                    # Track tool calls in the stream
+                    if "mult" in chunk.lower():
+                        if "mult" not in [call["tool"] for call in tool_calls_made]:
+                            tool_calls_made.append({"tool": "mult", "order": len(tool_calls_made) + 1})
+                            print(f"Tool call detected: mult (#{len(tool_calls_made)})")
+                    if "add" in chunk.lower():
+                        if "add" not in [call["tool"] for call in tool_calls_made]:
+                            tool_calls_made.append({"tool": "add", "order": len(tool_calls_made) + 1})
+                            print(f"Tool call detected: add (#{len(tool_calls_made)})")
+                    if "summarize" in chunk.lower():
+                        if "summarize_text" not in [call["tool"] for call in tool_calls_made]:
+                            tool_calls_made.append({"tool": "summarize_text", "order": len(tool_calls_made) + 1})
+                            print(f"Tool call detected: summarize_text (#{len(tool_calls_made)})")
+                    if "rephrase" in chunk.lower():
+                        if "rephrase_text" not in [call["tool"] for call in tool_calls_made]:
+                            tool_calls_made.append({"tool": "rephrase_text", "order": len(tool_calls_made) + 1})
+                            print(f"Tool call detected: rephrase_text (#{len(tool_calls_made)})")
+
+            response = await stream.aget_response()
+
+            print("="*50)
+            print(f"Streaming completed. Total deltas: {len(streaming_deltas)}")
+            print(f"Tool calls made: {[call['tool'] for call in tool_calls_made]}")
+            print(f"📄 Final response length: {len(response.response)} chars")
+            print(f"Final response: {response.response}")
+
+            # Validate tool usage sequence
+            tools_used = [call["tool"] for call in tool_calls_made]
+            print(f"🧪 Tools used in order: {tools_used}")
+
+            # Check that at least multiplication happened (basic requirement)
+            self.assertIn("mult", tools_used, f"Expected multiplication tool to be used. Tools used: {tools_used}")
+
+            # Check for mathematical results in the full response or streaming deltas
+            # Expected: 5*9=45, 45+13=58, 58*2=116
+            expected_intermediate_results = ["45", "58", "116"]
+            all_text = (full_response + " " + response.response).lower()
+            math_results_found = sum(1 for result in expected_intermediate_results
+                                     if result in all_text)
+
+            print(f"🔢 Mathematical results found: {math_results_found}/3 expected")
+            print(f"Full text searched: {all_text[:200]}...")
+
+            # More lenient assertion - just check that some mathematical progress was made
+            self.assertGreaterEqual(math_results_found, 1,
+                                    f"Expected at least 1 mathematical result. Found {math_results_found}. "
+                                    f"Full text: {all_text}")
+
+            # Verify that streaming actually produced content
+            self.assertGreater(len(streaming_deltas), 0, "Expected streaming deltas to be produced")
+            self.assertGreater(len(response.response.strip()), 0, "Expected non-empty final response")
+
+
+if __name__ == "__main__":
+    unittest.main()