vectara-agentic 0.1.9__tar.gz → 0.1.11__tar.gz

{vectara_agentic-0.1.9/vectara_agentic.egg-info → vectara_agentic-0.1.11}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: vectara_agentic
-Version: 0.1.9
+Version: 0.1.11
 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
@@ -16,27 +16,31 @@ Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: llama-index==0.11.10
+Requires-Dist: llama-index==0.11.13
 Requires-Dist: llama-index-indices-managed-vectara==0.2.2
 Requires-Dist: llama-index-agent-llm-compiler==0.2.0
-Requires-Dist: llama-index-agent-openai==0.3.1
-Requires-Dist: llama-index-llms-openai==0.2.7
+Requires-Dist: llama-index-agent-openai==0.3.4
+Requires-Dist: llama-index-llms-openai==0.2.9
 Requires-Dist: llama-index-llms-anthropic==0.3.1
 Requires-Dist: llama-index-llms-together==0.2.0
 Requires-Dist: llama-index-llms-groq==0.2.0
 Requires-Dist: llama-index-llms-fireworks==0.2.0
 Requires-Dist: llama-index-llms-cohere==0.3.0
+Requires-Dist: llama-index-llms-gemini==0.3.5
 Requires-Dist: llama-index-tools-yahoo-finance==0.2.0
 Requires-Dist: llama-index-tools-arxiv==0.2.0
 Requires-Dist: llama-index-tools-database==0.2.0
 Requires-Dist: llama-index-tools-google==0.2.0
 Requires-Dist: llama-index-tools-tavily_research==0.2.0
-Requires-Dist: llama-index-callbacks-arize-phoenix==0.2.1
-Requires-Dist: pydantic==2.8.2
+Requires-Dist: openinference-instrumentation-llama-index==3.0.2
+Requires-Dist: opentelemetry-proto==1.16.0
+Requires-Dist: arize-phoenix-otel==0.5.1
+Requires-Dist: pydantic==2.9.2
 Requires-Dist: retrying==1.3.4
-Requires-Dist: pymongo==4.6.1
+Requires-Dist: pymongo==4.6.3
 Requires-Dist: python-dotenv==1.0.1
 Requires-Dist: tiktoken==0.7.0
+Requires-Dist: dill==0.3.8
 
 # vectara-agentic
 
@@ -64,7 +68,7 @@ Requires-Dist: tiktoken==0.7.0
 - [Vectara account](https://console.vectara.com/signup/?utm_source=github&utm_medium=code&utm_term=DevRel&utm_content=vectara-agentic&utm_campaign=github-code-DevRel-vectara-agentic)
 - A Vectara corpus with an [API key](https://docs.vectara.com/docs/api-keys)
 - [Python 3.10 or higher](https://www.python.org/downloads/)
-- OpenAI API key (or API keys for Anthropic, TOGETHER.AI, Fireworks AI, Cohere or GROQ)
+- OpenAI API key (or API keys for Anthropic, TOGETHER.AI, Fireworks AI, Cohere, GEMINI or GROQ)
 
 ## Installation
 
@@ -98,6 +102,8 @@ query_financial_reports = vec_factory.create_rag_tool(
 )
 ```
 
+Note that `VECTARA_CORPUS_ID` can be a single ID or a comma-separated list of IDs.
+
 2. **Create other tools (optional)**
 
 In addition to RAG tools, you can generate a lot of other types of tools the agent can use. These could be mathematical tools, tools 
@@ -139,7 +145,7 @@ financial_bot_instructions = """
 Configure `vectara-agentic` using environment variables:
 
 - `VECTARA_AGENTIC_AGENT_TYPE`: valid values are `REACT`, `LLMCOMPILER` or `OPENAI` (default: `OPENAI`)
-- `VECTARA_AGENTIC_MAIN_LLM_PROVIDER`: valid values are `OPENAI`, `ANTHROPIC`, `TOGETHER`, `GROQ`, `COHERE` or `FIREWORKS` (default: `OPENAI`)
+- `VECTARA_AGENTIC_MAIN_LLM_PROVIDER`: valid values are `OPENAI`, `ANTHROPIC`, `TOGETHER`, `GROQ`, `COHERE`, `GEMINI` or `FIREWORKS` (default: `OPENAI`)
 - `VECTARA_AGENTIC_MAIN_MODEL_NAME`: agent model name (default depends on provider)
 - `VECTARA_AGENTIC_TOOL_LLM_PROVIDER`: tool LLM provider (default: `OPENAI`)
 - `VECTARA_AGENTIC_TOOL_MODEL_NAME`: tool model name (default depends on provider)
@@ -182,16 +188,27 @@ The `Agent` class defines a few helpful methods to help you understand the inter
 * The `report()` method prints out the agent object’s type, the tools, and the LLMs used for the main agent and tool calling.
 * The `token_counts()` method tells you how many tokens you have used in the current session for both the main agent and tool calling LLMs. This can be helpful if you want to track spend by token.
 
+## Serialization
+
+The `Agent` class supports serialization. Use the `dumps()` to serialize and `loads()` to read back from a serialized stream.
+
 ## Observability
 
 vectara-agentic supports observability via the existing integration of LlamaIndex and Arize Phoenix.
-To enable tracing of your vectara-agentic assistant, follow these steps (adapted from [here](https://docs.llamaindex.ai/en/stable/module_guides/observability/)):
-1. Go to `https://llamatrace.com/login` an create an account, then create an API key and put it in the `PHOENIX_API_KEY` variable
-2. `os["VECTARA_AGENTIC_OBSERVER_TYPE"] = "ARIZE_PHOENIX"`: to enable Arize Phoenix observability
-3. `os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = f"api_key={PHOENIX_API_KEY}"`
+First, set `os["VECTARA_AGENTIC_OBSERVER_TYPE"] = "ARIZE_PHOENIX"`.
+Then you can use Arize Phoenix in three ways: 
+1. **Locally**. 
+   1. If you have a local phoenix server that you've run using e.g. `python -m phoenix.server.main serve`, vectara-agentic will send all traces to it.
+   2. If not, vectara-agentic will run a local instance during the agent's lifecycle, and will close it when finished.
+   3. In both cases, traces will be sent to the local instance, and you can see the dashboard at `http://localhost:6006`
+2. **Hosted Instance**. In this case the traces are sent to the Phoenix instances hosted on Arize.
+   1. Go to `https://app.phoenix.arize.com`, setup an account if you don't have one.
+   2. create an API key and put it in the `PHOENIX_API_KEY` variable. This variable indicates you want to use the hosted version.
+   3. To view the traces go to `https://app.phoenix.arize.com`.
+
+Now when you run your agent, all call traces are sent to Phoenix and recorded. 
+In addition, vectara-agentic also records `FCS` values into Arize for every Vectara RAG call. You can see those results in the `Feedback` column of the arize UI.
 
-Now when you run your agent, all metrics are sent to LlamaTrace and recorded. You can view them at `https://llamatrace.com`.
-If you do not include the `OTEL_EXPORTER_OTLP_HEADERS` a local instance of Arize Phoenix will be setup instead and you can view it at `http://localhost:6006`
 
 ## About Custom Instructions
 

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/README.md

@@ -24,7 +24,7 @@
 - [Vectara account](https://console.vectara.com/signup/?utm_source=github&utm_medium=code&utm_term=DevRel&utm_content=vectara-agentic&utm_campaign=github-code-DevRel-vectara-agentic)
 - A Vectara corpus with an [API key](https://docs.vectara.com/docs/api-keys)
 - [Python 3.10 or higher](https://www.python.org/downloads/)
-- OpenAI API key (or API keys for Anthropic, TOGETHER.AI, Fireworks AI, Cohere or GROQ)
+- OpenAI API key (or API keys for Anthropic, TOGETHER.AI, Fireworks AI, Cohere, GEMINI or GROQ)
 
 ## Installation
 
@@ -58,6 +58,8 @@ query_financial_reports = vec_factory.create_rag_tool(
 )
 ```
 
+Note that `VECTARA_CORPUS_ID` can be a single ID or a comma-separated list of IDs.
+
 2. **Create other tools (optional)**
 
 In addition to RAG tools, you can generate a lot of other types of tools the agent can use. These could be mathematical tools, tools 
@@ -99,7 +101,7 @@ financial_bot_instructions = """
 Configure `vectara-agentic` using environment variables:
 
 - `VECTARA_AGENTIC_AGENT_TYPE`: valid values are `REACT`, `LLMCOMPILER` or `OPENAI` (default: `OPENAI`)
-- `VECTARA_AGENTIC_MAIN_LLM_PROVIDER`: valid values are `OPENAI`, `ANTHROPIC`, `TOGETHER`, `GROQ`, `COHERE` or `FIREWORKS` (default: `OPENAI`)
+- `VECTARA_AGENTIC_MAIN_LLM_PROVIDER`: valid values are `OPENAI`, `ANTHROPIC`, `TOGETHER`, `GROQ`, `COHERE`, `GEMINI` or `FIREWORKS` (default: `OPENAI`)
 - `VECTARA_AGENTIC_MAIN_MODEL_NAME`: agent model name (default depends on provider)
 - `VECTARA_AGENTIC_TOOL_LLM_PROVIDER`: tool LLM provider (default: `OPENAI`)
 - `VECTARA_AGENTIC_TOOL_MODEL_NAME`: tool model name (default depends on provider)
@@ -142,16 +144,27 @@ The `Agent` class defines a few helpful methods to help you understand the inter
 * The `report()` method prints out the agent object’s type, the tools, and the LLMs used for the main agent and tool calling.
 * The `token_counts()` method tells you how many tokens you have used in the current session for both the main agent and tool calling LLMs. This can be helpful if you want to track spend by token.
 
+## Serialization
+
+The `Agent` class supports serialization. Use the `dumps()` to serialize and `loads()` to read back from a serialized stream.
+
 ## Observability
 
 vectara-agentic supports observability via the existing integration of LlamaIndex and Arize Phoenix.
-To enable tracing of your vectara-agentic assistant, follow these steps (adapted from [here](https://docs.llamaindex.ai/en/stable/module_guides/observability/)):
-1. Go to `https://llamatrace.com/login` an create an account, then create an API key and put it in the `PHOENIX_API_KEY` variable
-2. `os["VECTARA_AGENTIC_OBSERVER_TYPE"] = "ARIZE_PHOENIX"`: to enable Arize Phoenix observability
-3. `os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = f"api_key={PHOENIX_API_KEY}"`
+First, set `os["VECTARA_AGENTIC_OBSERVER_TYPE"] = "ARIZE_PHOENIX"`.
+Then you can use Arize Phoenix in three ways: 
+1. **Locally**. 
+   1. If you have a local phoenix server that you've run using e.g. `python -m phoenix.server.main serve`, vectara-agentic will send all traces to it.
+   2. If not, vectara-agentic will run a local instance during the agent's lifecycle, and will close it when finished.
+   3. In both cases, traces will be sent to the local instance, and you can see the dashboard at `http://localhost:6006`
+2. **Hosted Instance**. In this case the traces are sent to the Phoenix instances hosted on Arize.
+   1. Go to `https://app.phoenix.arize.com`, setup an account if you don't have one.
+   2. create an API key and put it in the `PHOENIX_API_KEY` variable. This variable indicates you want to use the hosted version.
+   3. To view the traces go to `https://app.phoenix.arize.com`.
+
+Now when you run your agent, all call traces are sent to Phoenix and recorded. 
+In addition, vectara-agentic also records `FCS` values into Arize for every Vectara RAG call. You can see those results in the `Feedback` column of the arize UI.
 
-Now when you run your agent, all metrics are sent to LlamaTrace and recorded. You can view them at `https://llamatrace.com`.
-If you do not include the `OTEL_EXPORTER_OTLP_HEADERS` a local instance of Arize Phoenix will be setup instead and you can view it at `http://localhost:6006`
 
 ## About Custom Instructions
 

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/requirements.txt

@@ -1,21 +1,25 @@
-llama-index==0.11.10
+llama-index==0.11.13
 llama-index-indices-managed-vectara==0.2.2
 llama-index-agent-llm-compiler==0.2.0
-llama-index-agent-openai==0.3.1
-llama-index-llms-openai==0.2.7
+llama-index-agent-openai==0.3.4
+llama-index-llms-openai==0.2.9
 llama-index-llms-anthropic==0.3.1
 llama-index-llms-together==0.2.0
 llama-index-llms-groq==0.2.0
 llama-index-llms-fireworks==0.2.0
 llama-index-llms-cohere==0.3.0
+llama-index-llms-gemini==0.3.5
 llama-index-tools-yahoo-finance==0.2.0
 llama-index-tools-arxiv==0.2.0
 llama-index-tools-database==0.2.0
 llama-index-tools-google==0.2.0
 llama-index-tools-tavily_research==0.2.0
-llama-index-callbacks-arize-phoenix==0.2.1
-pydantic==2.8.2
+openinference-instrumentation-llama-index==3.0.2
+opentelemetry-proto==1.16.0
+arize-phoenix-otel==0.5.1
+pydantic==2.9.2
 retrying==1.3.4
-pymongo==4.6.1
+pymongo==4.6.3
 python-dotenv==1.0.1
 tiktoken==0.7.0
+dill==0.3.8

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/setup.py

@@ -8,7 +8,7 @@ def read_requirements():
 
 setup(
     name="vectara_agentic",
-    version="0.1.9",
+    version="0.1.11",
     author="Ofer Mendelevitch",
     author_email="ofer@vectara.com",
     description="A Python package for creating AI Assistants and AI Agents with Vectara",

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/tests/test_agent.py

@@ -56,6 +56,20 @@ class TestAgentPackage(unittest.TestCase):
         self.assertIsInstance(agent, Agent)
         self.assertEqual(agent._topic, "question answering")
 
+    def test_serialization(self):
+        agent = Agent.from_corpus(
+            tool_name="RAG Tool",
+            vectara_customer_id="4584783",
+            vectara_corpus_id="4",
+            vectara_api_key="api_key",
+            data_description="information",
+            assistant_specialty="question answering",
+        )
+
+        agent_reloaded = agent.loads(agent.dumps())
+        self.assertIsInstance(agent_reloaded, Agent)
+        self.assertEqual(agent, agent_reloaded)
+
 
 if __name__ == "__main__":
     unittest.main()

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/tests/test_tools.py

@@ -1,6 +1,7 @@
 import unittest
 
 from vectara_agentic.tools import VectaraTool, VectaraToolFactory, ToolsFactory, ToolType
+from vectara_agentic.agent import Agent
 from pydantic import Field, BaseModel
 from llama_index.core.tools import FunctionTool
 
@@ -57,6 +58,25 @@ class TestToolsPackage(unittest.TestCase):
         self.assertIsInstance(arxiv_tool, FunctionTool)
         self.assertEqual(arxiv_tool.tool_type, ToolType.QUERY)
 
+    def test_public_repo(self):
+        vectara_customer_id = "1366999410"
+        vectara_corpus_id = "1"
+        vectara_api_key = "zqt_UXrBcnI2UXINZkrv4g1tQPhzj02vfdtqYJIDiA"
+
+        class QueryToolArgs(BaseModel):
+            query: str = Field(description="The user query")
+
+        agent = Agent.from_corpus(
+            vectara_customer_id=vectara_customer_id,
+            vectara_corpus_id=vectara_corpus_id,
+            vectara_api_key=vectara_api_key,
+            tool_name="ask_vectara",
+            data_description="data from Vectara website",
+            assistant_specialty="RAG as a service",
+            vectara_summarizer="mockingbird-1.0-2024-07-16"
+        )
+
+        self.assertContains(agent.chat("What is Vectara?"), "Vectara is an end-to-end platform")
 
 if __name__ == "__main__":
     unittest.main()

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/vectara_agentic/__init__.py

@@ -3,7 +3,7 @@ vectara_agentic package.
 """
 
 # Define the package version
-__version__ = "0.1.9"
+__version__ = "0.1.11"
 
 # Import classes and functions from modules
 # from .module1 import Class1, function1

vectara_agentic-0.1.11/vectara_agentic/_observability.py

@@ -0,0 +1,83 @@
+import os
+import json
+import pandas as pd
+
+from phoenix.otel import register
+import phoenix as px
+from openinference.instrumentation.llama_index import LlamaIndexInstrumentor
+from phoenix.trace.dsl import SpanQuery
+from phoenix.trace import SpanEvaluations
+
+from .types import ObserverType
+
+def setup_observer():
+    observer = ObserverType(os.getenv("VECTARA_AGENTIC_OBSERVER_TYPE", "NO_OBSERVER"))
+    if observer == ObserverType.ARIZE_PHOENIX:
+        phoenix_endpoint = os.getenv("PHOENIX_ENDPOINT", None)
+        if not phoenix_endpoint:
+            px.launch_app()
+            tracer_provider = register(endpoint='http://localhost:6006', project_name="vectara-agentic")
+        elif 'app.phoenix.arize.com' in phoenix_endpoint:   # hosted on Arizze
+            phoenix_api_key = os.getenv("PHOENIX_API_KEY", None)
+            if not phoenix_api_key:
+                raise Exception("Arize Phoenix API key not set. Please set PHOENIX_API_KEY environment variable.")
+            os.environ["PHOENIX_CLIENT_HEADERS"] = f"api_key={phoenix_api_key}"
+            os.environ["PHOENIX_COLLECTOR_ENDPOINT"] = "https://app.phoenix.arize.com"
+            tracer_provider = register(endpoint=phoenix_endpoint, project_name="vectara-agentic")
+        else:       # Self hosted Phoenix
+            tracer_provider = register(endpoint=phoenix_endpoint, project_name="vectara-agentic")
+        LlamaIndexInstrumentor().instrument(tracer_provider=tracer_provider)
+    else:
+        print("No observer set.")
+
+
+def _extract_fcs_value(output):
+    try:
+        output_json = json.loads(output)
+        if 'metadata' in output_json and 'fcs' in output_json['metadata']:
+            return output_json['metadata']['fcs']
+    except json.JSONDecodeError:
+        print(f"Failed to parse JSON: {output}")
+    except KeyError:
+        print(f"'fcs' not found in: {output_json}")
+    return None
+
+def _find_top_level_parent_id(row, all_spans):
+    current_id = row['parent_id']
+    while current_id is not None:
+        parent_row = all_spans[all_spans.index == current_id]
+        if parent_row.empty:
+            break
+        new_parent_id = parent_row['parent_id'].iloc[0]
+        if new_parent_id == current_id:
+            break
+        if new_parent_id is None:
+            return current_id
+        current_id = new_parent_id
+    return current_id
+
+def eval_fcs():
+    query = SpanQuery().select(
+        "output.value",
+        "parent_id",
+        "name"
+    )
+    client = px.Client()
+    all_spans = client.query_spans(query, project_name="vectara-agentic")
+
+    vectara_spans = all_spans[all_spans['name'] == 'VectaraQueryEngine._query']
+    vectara_spans['top_level_parent_id'] = vectara_spans.apply(lambda row: _find_top_level_parent_id(row, all_spans), axis=1)
+    vectara_spans['score'] = vectara_spans['output.value'].apply(lambda x: _extract_fcs_value(x))
+    
+    vectara_spans.reset_index(inplace=True)
+    top_level_spans = vectara_spans.copy()
+    top_level_spans['context.span_id'] = top_level_spans['top_level_parent_id']
+    vectara_spans = pd.concat([vectara_spans, top_level_spans], ignore_index=True)
+    vectara_spans.set_index('context.span_id', inplace=True)
+
+    px.Client().log_evaluations(
+        SpanEvaluations(
+            dataframe=vectara_spans,
+            eval_name="Vectara FCS",
+        ),
+    )

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/vectara_agentic/agent.py

@@ -1,10 +1,12 @@
 """
 This module contains the Agent class for handling different types of agents and their interactions.
 """
-from typing import List, Callable, Optional
+from typing import List, Callable, Optional, Dict, Any
 import os
 from datetime import date
 import time
+import json
+import dill
 
 import logging
 logger = logging.getLogger('opentelemetry.exporter.otlp.proto.http.trace_exporter')
@@ -21,17 +23,16 @@ from llama_index.core.callbacks import CallbackManager, TokenCountingHandler
 from llama_index.core.callbacks.base_handler import BaseCallbackHandler
 from llama_index.agent.openai import OpenAIAgent
 from llama_index.core.memory import ChatMemoryBuffer
-from llama_index.core import set_global_handler
-
-import phoenix as px
 
 from dotenv import load_dotenv
 
-from .types import AgentType, AgentStatusType, LLMRole, ObserverType
+from .types import AgentType, AgentStatusType, LLMRole, ToolType
 from .utils import get_llm, get_tokenizer_for_model
 from ._prompts import REACT_PROMPT_TEMPLATE, GENERAL_PROMPT_TEMPLATE
 from ._callback import AgentCallbackHandler
-from .tools import VectaraToolFactory
+from ._observability import setup_observer, eval_fcs
+from .tools import VectaraToolFactory, VectaraTool
+
 
 load_dotenv(override=True)
 
@@ -67,7 +68,6 @@ class Agent:
     """
     Agent class for handling different types of agents and their interactions.
     """
-
     def __init__(
         self,
         tools: list[FunctionTool],
@@ -75,6 +75,7 @@ class Agent:
         custom_instructions: str = "",
         verbose: bool = True,
         update_func: Optional[Callable[[AgentStatusType, str], None]] = None,
+        agent_type: AgentType = None,
     ) -> None:
         """
         Initialize the agent with the specified type, tools, topic, and system message.
@@ -87,7 +88,7 @@ class Agent:
             verbose (bool, optional): Whether the agent should print its steps. Defaults to True.
             update_func (Callable): A callback function the code calls on any agent updates.
         """
-        self.agent_type = AgentType(os.getenv("VECTARA_AGENTIC_AGENT_TYPE", "OPENAI"))
+        self.agent_type = agent_type or AgentType(os.getenv("VECTARA_AGENTIC_AGENT_TYPE", "OPENAI"))
         self.tools = tools
         self.llm = get_llm(LLMRole.MAIN)
         self._custom_instructions = custom_instructions
@@ -140,17 +141,49 @@ class Agent:
         else:
             raise ValueError(f"Unknown agent type: {self.agent_type}")
 
-        observer = ObserverType(os.getenv("VECTARA_AGENTIC_OBSERVER_TYPE", "NO_OBSERVER"))
-        if observer == ObserverType.ARIZE_PHOENIX:
-            if os.environ.get("OTEL_EXPORTER_OTLP_HEADERS", None):
-                set_global_handler("arize_phoenix", endpoint="https://llamatrace.com/v1/traces")
-                print("Arize Phoenix observer set. https://llamatrace.com")
-            else:
-                px.launch_app()
-                set_global_handler("arize_phoenix", endpoint="http://localhost:6006/v1/traces")
-                print("Arize Phoenix observer set. http://localhost:6006/.")
-        else:
-            print("No observer set.")
+        try:
+            setup_observer()
+        except Exception as e:
+            print(f"Failed to set up observer ({e}), ignoring")
+
+    def __eq__(self, other):
+        if not isinstance(other, Agent):
+            print(f"Comparison failed: other is not an instance of Agent. (self: {type(self)}, other: {type(other)})")
+            return False
+
+        # Compare agent_type
+        if self.agent_type != other.agent_type:
+            print(f"Comparison failed: agent_type differs. (self.agent_type: {self.agent_type}, other.agent_type: {other.agent_type})")
+            return False
+
+        # Compare tools
+        if self.tools != other.tools:
+            print(f"Comparison failed: tools differ. (self.tools: {self.tools}, other.tools: {other.tools})")
+            return False
+
+        # Compare topic
+        if self._topic != other._topic:
+            print(f"Comparison failed: topic differs. (self.topic: {self._topic}, other.topic: {other._topic})")
+            return False
+
+        # Compare custom_instructions
+        if self._custom_instructions != other._custom_instructions:
+            print(f"Comparison failed: custom_instructions differ. (self.custom_instructions: {self._custom_instructions}, other.custom_instructions: {other._custom_instructions})")
+            return False
+
+        # Compare verbose
+        if self.verbose != other.verbose:
+            print(f"Comparison failed: verbose differs. (self.verbose: {self.verbose}, other.verbose: {other.verbose})")
+            return False
+
+        # Compare agent
+        if self.agent.memory.chat_store != other.agent.memory.chat_store:
+            print(f"Comparison failed: agent memory differs. (self.agent: {repr(self.agent.memory.chat_store)}, other.agent: {repr(other.agent.memory.chat_store)})")
+            return False
+
+        # If all comparisons pass
+        print("All comparisons passed. Objects are equal.")
+        return True
 
     @classmethod
     def from_tools(
@@ -160,6 +193,7 @@ class Agent:
         custom_instructions: str = "",
         verbose: bool = True,
         update_func: Optional[Callable[[AgentStatusType, str], None]] = None,
+        agent_type: AgentType = None,
     ) -> "Agent":
         """
         Create an agent from tools, agent type, and language model.
@@ -176,7 +210,7 @@ class Agent:
         Returns:
             Agent: An instance of the Agent class.
         """
-        return cls(tools, topic, custom_instructions, verbose, update_func)
+        return cls(tools, topic, custom_instructions, verbose, update_func, agent_type)
 
     @classmethod
     def from_corpus(
@@ -203,12 +237,12 @@ class Agent:
         Args:
             tool_name (str): The name of Vectara tool used by the agent
             vectara_customer_id (str): The Vectara customer ID.
-            vectara_corpus_id (str): The Vectara corpus ID.
+            vectara_corpus_id (str): The Vectara corpus ID (or comma separated list of IDs).
             vectara_api_key (str): The Vectara API key.
             data_description (str): The description of the data.
             assistant_specialty (str): The specialty of the assistant.
             verbose (bool, optional): Whether to print verbose output.
-            vectara_filter_fields (List[dict], optional): The filterable attributes (each dict includes name, type, and description).
+            vectara_filter_fields (List[dict], optional): The filterable attributes (each dict maps field name to Tuple[type, description]).
             vectara_lambda_val (float, optional): The lambda value for Vectara hybrid search.
             vectara_reranker (str, optional): The Vectara reranker name (default "mmr")
             vectara_rerank_k (int, optional): The number of results to use with reranking.
@@ -224,9 +258,10 @@ class Agent:
                                          vectara_customer_id=vectara_customer_id,
                                          vectara_corpus_id=vectara_corpus_id)
         field_definitions = {}
-        field_definitions['query'] = (str, Field(description="The user query"))
+        field_definitions['query'] = (str, Field(description="The user query"))  # type: ignore
         for field in vectara_filter_fields:
-            field_definitions[field['name']] = (eval(field['type']), Field(description=field['description'], default=None))  # type: ignore
+            field_definitions[field['name']] = (eval(field['type']), 
+                                                Field(description=field['description']))  # type: ignore
         QueryArgs = create_model(   # type: ignore
             "QueryArgs",
             **field_definitions
@@ -311,7 +346,126 @@ class Agent:
             agent_response = self.agent.chat(prompt)
             if self.verbose:
                 print(f"Time taken: {time.time() - st}")
+            eval_fcs()
             return agent_response.response
         except Exception as e:
             import traceback
             return f"Vectara Agentic: encountered an exception ({e}) at ({traceback.format_exc()}), and can't respond."
+
+    # Serialization methods
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Serialize the Agent instance to a dictionary."""
+        tool_info = []
+
+        for tool in self.tools:
+            tool_dict = {
+                "tool_type": tool.tool_type.value,
+                "name": tool._metadata.name,
+                "description": tool._metadata.description,
+                "fn": dill.dumps(tool.fn).decode('latin-1') if tool.fn else None,  # Serialize fn
+                "async_fn": dill.dumps(tool.async_fn).decode('latin-1') if tool.async_fn else None,  # Serialize async_fn
+            }
+            tool_info.append(tool_dict)
+
+        return {
+            "agent_type": self.agent_type.value,
+            "tools": tool_info,
+            "topic": self._topic,
+            "custom_instructions": self._custom_instructions,
+            "verbose": self.verbose,
+        }
+
+
+    def dumps(self) -> str:
+        """Serialize the Agent instance to a JSON string."""
+        return json.dumps(self.to_dict())
+
+    @classmethod
+    def loads(cls, data: str) -> "Agent":
+        """Create an Agent instance from a JSON string."""
+        return cls.from_dict(json.loads(data))
+
+
+    def to_dict(self) -> Dict[str, Any]:
+        """Serialize the Agent instance to a dictionary."""
+        tool_info = []
+
+        for tool in self.tools:
+            # Serialize each tool's metadata, function, and dynamic model schema (QueryArgs)
+            tool_dict = {
+                "tool_type": tool.tool_type.value,
+                "name": tool._metadata.name,
+                "description": tool._metadata.description,
+                "fn": dill.dumps(tool.fn).decode('latin-1') if tool.fn else None,  # Serialize fn
+                "async_fn": dill.dumps(tool.async_fn).decode('latin-1') if tool.async_fn else None,  # Serialize async_fn
+                "fn_schema": tool._metadata.fn_schema.model_json_schema() if hasattr(tool._metadata, 'fn_schema') else None,  # Serialize schema if available
+            }
+            tool_info.append(tool_dict)
+
+        return {
+            "agent_type": self.agent_type.value,
+            "memory": dill.dumps(self.agent.memory).decode('latin-1'),
+            "tools": tool_info,
+            "topic": self._topic,
+            "custom_instructions": self._custom_instructions,
+            "verbose": self.verbose,
+        }
+
+    @classmethod
+    def from_dict(cls, data: Dict[str, Any]) -> "Agent":
+        """Create an Agent instance from a dictionary."""
+        agent_type = AgentType(data["agent_type"])
+        tools = []
+
+        JSON_TYPE_TO_PYTHON = {
+            "string": "str",
+            "integer": "int",
+            "boolean": "bool",
+            "array": "list",
+            "object": "dict",
+            "number": "float",
+        }
+    
+        for tool_data in data["tools"]:
+            # Recreate the dynamic model using the schema info
+            if tool_data.get("fn_schema"):
+                field_definitions = {}
+                for field,values in tool_data["fn_schema"]["properties"].items():
+                    if 'default' in values:
+                        field_definitions[field] = (eval(JSON_TYPE_TO_PYTHON.get(values['type'], values['type'])), 
+                                                             Field(description=values['description'], default=values['default']))  # type: ignore
+                    else:
+                        field_definitions[field] = (eval(JSON_TYPE_TO_PYTHON.get(values['type'], values['type'])), 
+                                                            Field(description=values['description']))    # type: ignore
+                query_args_model = create_model(   # type: ignore
+                    "QueryArgs",
+                    **field_definitions
+                )
+            else:
+                query_args_model = create_model("QueryArgs")
+
+            fn = dill.loads(tool_data["fn"].encode('latin-1')) if tool_data["fn"] else None
+            async_fn = dill.loads(tool_data["async_fn"].encode('latin-1')) if tool_data["async_fn"] else None
+
+            tool = VectaraTool.from_defaults(
+                tool_type=ToolType(tool_data["tool_type"]),
+                name=tool_data["name"],
+                description=tool_data["description"],
+                fn=fn,
+                async_fn=async_fn,
+                fn_schema=query_args_model  # Re-assign the recreated dynamic model
+            )
+            tools.append(tool)
+
+        agent = cls(
+            tools=tools,
+            agent_type=agent_type,
+            topic=data["topic"],
+            custom_instructions=data["custom_instructions"],
+            verbose=data["verbose"],
+        )
+        memory = dill.loads(data["memory"].encode('latin-1')) if data.get("memory") else None
+        if memory:
+            agent.agent.memory = memory
+        return agent

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/vectara_agentic/tools.py

@@ -81,12 +81,31 @@ class VectaraTool(FunctionTool):
         tool = FunctionTool.from_defaults(fn, name, description, return_direct, fn_schema, async_fn, tool_metadata)
         vectara_tool = cls(
             tool_type=tool_type,
-            fn=tool.fn,
-            metadata=tool.metadata,
-            async_fn=tool.async_fn
+            fn=tool._fn,
+            metadata=tool._metadata,
+            async_fn=tool._async_fn
         )
         return vectara_tool
 
+    def __eq__(self, other):
+        if self.tool_type != other.tool_type:
+            return False
+
+        # Check if fn_schema is an instance of a BaseModel or a class itself (metaclass)
+        self_schema_dict = self.metadata.fn_schema.model_fields
+        other_schema_dict = other.metadata.fn_schema.model_fields
+        is_equal = True
+        for key in self_schema_dict.keys():
+            if key not in other_schema_dict:
+                is_equal = False
+                break
+            if (self_schema_dict[key].annotation != other_schema_dict[key].annotation or
+                self_schema_dict[key].description != other_schema_dict[key].description or
+                self_schema_dict[key].is_required() != other_schema_dict[key].is_required()):
+                is_equal = False
+                break
+        return is_equal
+
 
 class VectaraToolFactory:
     """
@@ -103,12 +122,13 @@ class VectaraToolFactory:
         Initialize the VectaraToolFactory
         Args:
             vectara_customer_id (str): The Vectara customer ID.
-            vectara_corpus_id (str): The Vectara corpus ID.
+            vectara_corpus_id (str): The Vectara corpus ID (or comma separated list of IDs).
             vectara_api_key (str): The Vectara API key.
         """
         self.vectara_customer_id = vectara_customer_id
         self.vectara_corpus_id = vectara_corpus_id
         self.vectara_api_key = vectara_api_key
+        self.num_corpora = len(vectara_corpus_id.split(","))
 
     def create_rag_tool(
         self,
@@ -188,7 +208,7 @@ class VectaraToolFactory:
                 summary_response_lang=summary_response_lang,
                 summary_prompt_name=vectara_summarizer,
                 reranker=reranker,
-                rerank_k=rerank_k,
+                rerank_k=rerank_k if rerank_k*self.num_corpora<=100 else int(100/self.num_corpora),
                 mmr_diversity_bias=mmr_diversity_bias,
                 n_sentence_before=n_sentences_before,
                 n_sentence_after=n_sentences_after,
@@ -255,7 +275,7 @@ class VectaraToolFactory:
             )
             return out
 
-        fields = tool_args_schema.__fields__
+        fields = tool_args_schema.model_fields
         params = [
             inspect.Parameter(
                 name=field_name,

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/vectara_agentic/types.py

@@ -28,6 +28,7 @@ class ModelProvider(Enum):
     GROQ = "GROQ"
     FIREWORKS = "FIREWORKS"
     COHERE = "COHERE"
+    GEMINI = "GEMINI"
 
 
 class AgentStatusType(Enum):

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/vectara_agentic/utils.py

@@ -11,6 +11,7 @@ from llama_index.llms.together import TogetherLLM
 from llama_index.llms.groq import Groq
 from llama_index.llms.fireworks import Fireworks
 from llama_index.llms.cohere import Cohere
+from llama_index.llms.gemini import Gemini
 
 import tiktoken
 from typing import Tuple, Callable, Optional
@@ -24,6 +25,7 @@ provider_to_default_model_name = {
     ModelProvider.GROQ: "llama-3.1-70b-versatile",
     ModelProvider.FIREWORKS: "accounts/fireworks/models/firefunction-v2",
     ModelProvider.COHERE: "command-r-plus",
+    ModelProvider.GEMINI: "models/gemini-pro",
 }
 
 DEFAULT_MODEL_PROVIDER = ModelProvider.OPENAI
@@ -77,17 +79,19 @@ def get_llm(role: LLMRole) -> LLM:
     model_provider, model_name = _get_llm_params_for_role(role)
 
     if model_provider == ModelProvider.OPENAI:
-        llm = OpenAI(model=model_name, temperature=0)
+        llm = OpenAI(model=model_name, temperature=0, is_function_calling_model=True)
     elif model_provider == ModelProvider.ANTHROPIC:
-        llm = Anthropic(model=model_name, temperature=0)
+        llm = Anthropic(model=model_name, temperature=0, is_function_calling_model=True)
+    elif model_provider == ModelProvider.GEMINI:
+        llm = Gemini(model=model_name, temperature=0, is_function_calling_model=True)
     elif model_provider == ModelProvider.TOGETHER:
-        llm = TogetherLLM(model=model_name, temperature=0)
+        llm = TogetherLLM(model=model_name, temperature=0, is_function_calling_model=True)
     elif model_provider == ModelProvider.GROQ:
-        llm = Groq(model=model_name, temperature=0)
+        llm = Groq(model=model_name, temperature=0, is_function_calling_model=True)
     elif model_provider == ModelProvider.FIREWORKS:
-        llm = Fireworks(model=model_name, temperature=0)
+        llm = Fireworks(model=model_name, temperature=0, is_function_calling_model=True)
     elif model_provider == ModelProvider.COHERE:
-        llm = Cohere(model=model_name, temperature=0)
+        llm = Cohere(model=model_name, temperature=0, is_function_calling_model=True)
     else:
         raise ValueError(f"Unknown LLM provider: {model_provider}")
 

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11/vectara_agentic.egg-info}/PKG-INFO

@@ -1,6 +1,6 @@
 Metadata-Version: 2.1
 Name: vectara_agentic
-Version: 0.1.9
+Version: 0.1.11
 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
@@ -16,27 +16,31 @@ Classifier: Topic :: Software Development :: Libraries :: Python Modules
 Requires-Python: >=3.10
 Description-Content-Type: text/markdown
 License-File: LICENSE
-Requires-Dist: llama-index==0.11.10
+Requires-Dist: llama-index==0.11.13
 Requires-Dist: llama-index-indices-managed-vectara==0.2.2
 Requires-Dist: llama-index-agent-llm-compiler==0.2.0
-Requires-Dist: llama-index-agent-openai==0.3.1
-Requires-Dist: llama-index-llms-openai==0.2.7
+Requires-Dist: llama-index-agent-openai==0.3.4
+Requires-Dist: llama-index-llms-openai==0.2.9
 Requires-Dist: llama-index-llms-anthropic==0.3.1
 Requires-Dist: llama-index-llms-together==0.2.0
 Requires-Dist: llama-index-llms-groq==0.2.0
 Requires-Dist: llama-index-llms-fireworks==0.2.0
 Requires-Dist: llama-index-llms-cohere==0.3.0
+Requires-Dist: llama-index-llms-gemini==0.3.5
 Requires-Dist: llama-index-tools-yahoo-finance==0.2.0
 Requires-Dist: llama-index-tools-arxiv==0.2.0
 Requires-Dist: llama-index-tools-database==0.2.0
 Requires-Dist: llama-index-tools-google==0.2.0
 Requires-Dist: llama-index-tools-tavily_research==0.2.0
-Requires-Dist: llama-index-callbacks-arize-phoenix==0.2.1
-Requires-Dist: pydantic==2.8.2
+Requires-Dist: openinference-instrumentation-llama-index==3.0.2
+Requires-Dist: opentelemetry-proto==1.16.0
+Requires-Dist: arize-phoenix-otel==0.5.1
+Requires-Dist: pydantic==2.9.2
 Requires-Dist: retrying==1.3.4
-Requires-Dist: pymongo==4.6.1
+Requires-Dist: pymongo==4.6.3
 Requires-Dist: python-dotenv==1.0.1
 Requires-Dist: tiktoken==0.7.0
+Requires-Dist: dill==0.3.8
 
 # vectara-agentic
 
@@ -64,7 +68,7 @@ Requires-Dist: tiktoken==0.7.0
 - [Vectara account](https://console.vectara.com/signup/?utm_source=github&utm_medium=code&utm_term=DevRel&utm_content=vectara-agentic&utm_campaign=github-code-DevRel-vectara-agentic)
 - A Vectara corpus with an [API key](https://docs.vectara.com/docs/api-keys)
 - [Python 3.10 or higher](https://www.python.org/downloads/)
-- OpenAI API key (or API keys for Anthropic, TOGETHER.AI, Fireworks AI, Cohere or GROQ)
+- OpenAI API key (or API keys for Anthropic, TOGETHER.AI, Fireworks AI, Cohere, GEMINI or GROQ)
 
 ## Installation
 
@@ -98,6 +102,8 @@ query_financial_reports = vec_factory.create_rag_tool(
 )
 ```
 
+Note that `VECTARA_CORPUS_ID` can be a single ID or a comma-separated list of IDs.
+
 2. **Create other tools (optional)**
 
 In addition to RAG tools, you can generate a lot of other types of tools the agent can use. These could be mathematical tools, tools 
@@ -139,7 +145,7 @@ financial_bot_instructions = """
 Configure `vectara-agentic` using environment variables:
 
 - `VECTARA_AGENTIC_AGENT_TYPE`: valid values are `REACT`, `LLMCOMPILER` or `OPENAI` (default: `OPENAI`)
-- `VECTARA_AGENTIC_MAIN_LLM_PROVIDER`: valid values are `OPENAI`, `ANTHROPIC`, `TOGETHER`, `GROQ`, `COHERE` or `FIREWORKS` (default: `OPENAI`)
+- `VECTARA_AGENTIC_MAIN_LLM_PROVIDER`: valid values are `OPENAI`, `ANTHROPIC`, `TOGETHER`, `GROQ`, `COHERE`, `GEMINI` or `FIREWORKS` (default: `OPENAI`)
 - `VECTARA_AGENTIC_MAIN_MODEL_NAME`: agent model name (default depends on provider)
 - `VECTARA_AGENTIC_TOOL_LLM_PROVIDER`: tool LLM provider (default: `OPENAI`)
 - `VECTARA_AGENTIC_TOOL_MODEL_NAME`: tool model name (default depends on provider)
@@ -182,16 +188,27 @@ The `Agent` class defines a few helpful methods to help you understand the inter
 * The `report()` method prints out the agent object’s type, the tools, and the LLMs used for the main agent and tool calling.
 * The `token_counts()` method tells you how many tokens you have used in the current session for both the main agent and tool calling LLMs. This can be helpful if you want to track spend by token.
 
+## Serialization
+
+The `Agent` class supports serialization. Use the `dumps()` to serialize and `loads()` to read back from a serialized stream.
+
 ## Observability
 
 vectara-agentic supports observability via the existing integration of LlamaIndex and Arize Phoenix.
-To enable tracing of your vectara-agentic assistant, follow these steps (adapted from [here](https://docs.llamaindex.ai/en/stable/module_guides/observability/)):
-1. Go to `https://llamatrace.com/login` an create an account, then create an API key and put it in the `PHOENIX_API_KEY` variable
-2. `os["VECTARA_AGENTIC_OBSERVER_TYPE"] = "ARIZE_PHOENIX"`: to enable Arize Phoenix observability
-3. `os.environ["OTEL_EXPORTER_OTLP_HEADERS"] = f"api_key={PHOENIX_API_KEY}"`
+First, set `os["VECTARA_AGENTIC_OBSERVER_TYPE"] = "ARIZE_PHOENIX"`.
+Then you can use Arize Phoenix in three ways: 
+1. **Locally**. 
+   1. If you have a local phoenix server that you've run using e.g. `python -m phoenix.server.main serve`, vectara-agentic will send all traces to it.
+   2. If not, vectara-agentic will run a local instance during the agent's lifecycle, and will close it when finished.
+   3. In both cases, traces will be sent to the local instance, and you can see the dashboard at `http://localhost:6006`
+2. **Hosted Instance**. In this case the traces are sent to the Phoenix instances hosted on Arize.
+   1. Go to `https://app.phoenix.arize.com`, setup an account if you don't have one.
+   2. create an API key and put it in the `PHOENIX_API_KEY` variable. This variable indicates you want to use the hosted version.
+   3. To view the traces go to `https://app.phoenix.arize.com`.
+
+Now when you run your agent, all call traces are sent to Phoenix and recorded. 
+In addition, vectara-agentic also records `FCS` values into Arize for every Vectara RAG call. You can see those results in the `Feedback` column of the arize UI.
 
-Now when you run your agent, all metrics are sent to LlamaTrace and recorded. You can view them at `https://llamatrace.com`.
-If you do not include the `OTEL_EXPORTER_OTLP_HEADERS` a local instance of Arize Phoenix will be setup instead and you can view it at `http://localhost:6006`
 
 ## About Custom Instructions
 

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/vectara_agentic.egg-info/SOURCES.txt

@@ -7,6 +7,7 @@ tests/test_agent.py
 tests/test_tools.py
 vectara_agentic/__init__.py
 vectara_agentic/_callback.py
+vectara_agentic/_observability.py
 vectara_agentic/_prompts.py
 vectara_agentic/agent.py
 vectara_agentic/tools.py

{vectara_agentic-0.1.9 → vectara_agentic-0.1.11}/vectara_agentic.egg-info/requires.txt

@@ -1,21 +1,25 @@
-llama-index==0.11.10
+llama-index==0.11.13
 llama-index-indices-managed-vectara==0.2.2
 llama-index-agent-llm-compiler==0.2.0
-llama-index-agent-openai==0.3.1
-llama-index-llms-openai==0.2.7
+llama-index-agent-openai==0.3.4
+llama-index-llms-openai==0.2.9
 llama-index-llms-anthropic==0.3.1
 llama-index-llms-together==0.2.0
 llama-index-llms-groq==0.2.0
 llama-index-llms-fireworks==0.2.0
 llama-index-llms-cohere==0.3.0
+llama-index-llms-gemini==0.3.5
 llama-index-tools-yahoo-finance==0.2.0
 llama-index-tools-arxiv==0.2.0
 llama-index-tools-database==0.2.0
 llama-index-tools-google==0.2.0
 llama-index-tools-tavily_research==0.2.0
-llama-index-callbacks-arize-phoenix==0.2.1
-pydantic==2.8.2
+openinference-instrumentation-llama-index==3.0.2
+opentelemetry-proto==1.16.0
+arize-phoenix-otel==0.5.1
+pydantic==2.9.2
 retrying==1.3.4
-pymongo==4.6.1
+pymongo==4.6.3
 python-dotenv==1.0.1
 tiktoken==0.7.0
+dill==0.3.8