SQLAgent Community

SQLAgent Community is a Python library for creating intelligent agents that can query SQLite databases using LLMs, generate SQL from natural language, and maintain conversation memory. Ideal for exploratory data analysis, conversational dashboards, or automating complex queries.

Requirements

• Python >= 3.10
• SQLite
• Ollama installed and configured
  • Ollama provides the LLM and embedding models used by SQLAgent
• Pandas (for reading Excel files)
• Rich (optional, for nice logs)

Installation

You can install via pip (once published on PyPI):

pip install sqlagent-community

Configuration

Suggested file structure

dataset/
├─ data.db                  # SQLite database
├─ excels/                   # Excel files to build the DB
vector_db/
├─ schema_embeddings.pkl     # Schema embeddings
context/
├─ file.md                   # Additional context for the agent

Initialize Builder and database

from sqlagent_community.Builder import SQLBuilder

builder = SQLBuilder(
    excel_dir="dataset/excels",
    db_file="dataset/data.db",
    embed_file="vector_db/schema_embeddings.pkl",
    embed_model="qwen3-embedding:0.6b",
    verbose=True
)

builder.build()

Set up Chat and Memory

from sqlagent_community.memory.ChatManager import ChatManager
from sqlagent_community.memory.ConversationSummaryBufferMemory import ConversationSummaryBufferMemory

manager = ChatManager()
chat_id = manager.create_chat()

memory = ConversationSummaryBufferMemory(chat_id=chat_id)

Initialize the Agent

from sqlagent_community.Agent import SQLAgent

agent = SQLAgent(
    db_path="dataset/data.db",
    embed_file="vector_db/schema_embeddings.pkl",
    embedding_model="qwen3-embedding:0.6b",
    llm_model="qwen2.5:7b",
    text2sql="qwen2.5-coder:7b",
    memory=memory,
    context="./context/file.md",
    verbose=True
)

Asking questions to the agent

question = "What is the average altitude per municipality of the collections?"
result = agent.run(question)

print(result)

The agent converts the question into SQL, executes it on the database, and returns the result, while maintaining conversation history and context.

Features

• Automatic SQLite database construction from Excel files.
• SQL generation via LLM from natural language.
• Summarized conversational memory per chat.
• Additional context from Markdown files.
• Detailed logs for debugging.

License

MIT License – see the LICENSE file for details.