Agents & Examples

Agentic AI Kit provides a collection of specialized agents for conversational AI, reasoning, analytics, retrieval, structured outputs, and automated machine learning. All agents share a consistent interface while extending the capabilities of BaseAgent.

BaseAgent

File: agentic_ai/agents/base.py

The foundation of the framework. BaseAgent supports conversational AI with configurable Gemini models, memory management, streaming responses, and optional long-term fact extraction.

Features

Persona and system prompts
Short-term conversation memory
Durable fact extraction
Streaming and non-streaming generation
Configurable Gemini models

Constructor

BaseAgent( name, sys_prompt, model="gemini-2.5-flash-lite", ...)

Example

from agentic_ai.agents import BaseAgent
agent = BaseAgent( name="Sage", sys_prompt="You are a concise Python tutor.")
response = agent.think( "Explain list comprehensions.", stream=False,)
print(response)
ToolAgent

File: agentic_ai/agents/tool_agent.py

Extends BaseAgent with Gemini function calling. Register Python functions and allow the model to invoke them automatically when appropriate.

Features

Automatic tool selection
Gemini function calling
Python function registration
Parameter descriptions
Structured tool schemas

Example

from agentic_ai.agents import ToolAgent
def area(length, width):    return length * width
agent = ToolAgent(    name="Calculator",    sys_prompt="Use tools for calculations.")
agent.register_tool(area)
print(    agent.think(        "Area of a 12 by 8 rectangle?"    ))
JsonAgent

File: agentic_ai/agents/json_agent.py

Produces structured Python dictionaries instead of free-form text.

Ideal for:

information extraction
APIs
automation
structured workflows

Example

agent = JsonAgent( name="Extractor", schema={ "people":["string"], "places":["string"] })
result = agent.think( "Alice visited Paris.")
print(result)

ReasoningAgent

File: agentic_ai/agents/reasoning_agent.py

Generates an explicit reasoning plan together with the final answer.

Returns:

reasoning
answer
context_for_next

Useful for planning, orchestration and multi-agent workflows.

Example

planner = ReasoningAgent( name="Planner", sys_prompt="Break tasks into steps.")
result = planner.think( "Plan a product launch.", stream=False)
print(result.reasoning)print(result.answer)

AnalystAgent

File: agentic_ai/agents/analyst_agent.py

A dataframe-aware analytics specialist.

It combines local pandas analysis with Gemini-powered interpretation.

Capabilities

Data profiling
Numeric summaries
Categorical summaries
Grouped aggregations
Correlation analysis
Natural-language insights

Example

agent = AnalystAgent()
analysis = agent.analyze_dataframe( df, question="What are the major business insights?")
print(analysis)

MLEAgent

File: agentic_ai/agents/mle_agent.py

Automates classical machine-learning workflows using scikit-learn.

Workflow

Profile dataset
Summarize features
Detect leakage
Train baseline models
Compare performance
Select best pipeline
Optional Gemini interpretation

Supported Models

Classification

Logistic Regression
Random Forest Classifier

Regression

Linear Regression
Random Forest Regressor

Example

agent = MLEAgent()
result = agent.create_model( df=df, target_col="target")
print(result["best_model"])

AutoModelAgent

File: agentic_ai/agents/auto_model_agent.py

Extends MLEAgent into a complete end-to-end modelling pipeline.

Workflow

Profile data
Analyse target
Summarise features
Detect leakage
Generate modelling policy
Train models
Select best pipeline
Interpret results

Example

agent = AutoModelAgent()
result = agent.run( df=df, target_col="sale_price")
print(result["best_model"])

RAGAgent

File: agentic_ai/rag/rag_agent.py

Retrieval-Augmented Generation with built-in chunking, embeddings and vector search.

Features

Text chunking
Gemini embeddings
In-memory vector database
Semantic retrieval
Context-grounded generation

Example

agent = RAGAgent( name="Policy Assistant", top_k=3)
agent.ingest(document)
answer = agent.think( "What is the refund policy?")
print(answer)

Agent Comparison

Agent Best For

BaseAgent > Conversational AI

ToolAgent > Function calling

JsonAgent > Structured outputs

ReasoningAgent > Planning workflows

AnalystAgent > Data analysis

MLEAgent > Classical ML

AutoModelAgent > Automated modelling

RAGAgent > Retrieval-Augmented Generation

Page updated

Google Sites

Report abuse