query3ai 0.2.1

Query3AI: A multi-agent system combining document structure extraction, relevance filtering, and reasoning with Neo4j.

Query3AI

An intelligent, local-first document query system powered by a 3-Agent AI pipeline and Neo4j graph storage.

---

What Is Query3AI?

Query3AI lets you ingest documents (PDF, DOCX, TXT, MD) and query them in natural language. It does not flatten your documents into a pile of text chunks like standard AI tools. It reads the structure, builds a knowledge graph, and reasons with three specialised AI agents — one to organise, one to filter, one to answer.

# Install once
pip install query3ai

# Initialize workspace (creates ~/.query3ai with config)
query3ai init

# Start Neo4j
query3ai start-db

# Ingest documents
query3ai ingest report.pdf

# Query with interactive chat
query3ai chat

No cloud required. No API keys needed for local models. Runs on a standard laptop.

---

Installation

From PyPI (Recommended)

pip install query3ai

From Source

git clone https://github.com/vivekvpai/Query3AI.git
cd Query3AI
pip install -e .

---

Quick Start

1. Initialize Query3AI

# Global workspace (default - stores config in ~/.query3ai)
query3ai init

# Or local workspace (creates config in current directory)
query3ai init --local

This creates:

• ~/.query3ai/docker-compose.yml - Neo4j configuration
• ~/.query3ai/.env - Environment variables
• ~/.query3ai/config.json - Model configuration

2. Start Neo4j

query3ai start-db

Or manually with Docker:

docker run -p 7687:7687 -p 7474:7474 \
  -e NEO4J_AUTH=neo4j/query3ai \
  neo4j:latest

Default connection:

• URI: bolt://localhost:7687
• User: neo4j
• Password: query3ai

3. Start Ollama (Optional - for local models)

# Make sure Ollama is running
ollama serve

# Pull the three agent models
ollama pull phi3.5          # Tree Agent
ollama pull gemma2:2b       # Decision Agent
ollama pull deepseek-r1:7b  # Reasoning Agent

4. Ingest and Query

# Ingest a document
query3ai ingest path/to/document.pdf

# Start interactive chat
query3ai chat

---

CLI Commands

Command	Description
query3ai init	Initialize workspace in ~/.query3ai
query3ai init --local	Initialize workspace in current directory
query3ai start-db	Start Neo4j via docker-compose
query3ai stop-db	Stop Neo4j
query3ai ingest <file>	Ingest a PDF, DOCX, TXT, or MD file
query3ai ask "<question>"	Query all ingested documents
query3ai ask "<question>" --cloud	Query using cloud models
query3ai list	List all ingested documents
query3ai inspect <doc_id>	Inspect a document's tree structure
query3ai delete <doc_id>	Delete a document and all its nodes
query3ai chat	Start interactive TUI chat

---

The 3-Agent Pipeline

Document
    │
    ▼
[Agent 1 — Tree AI]      phi3.5 / qwen3.5:cloud
Builds hierarchical tree: Document → Sections → Chunks
    │
    ▼
Neo4j Graph Database
    │
    ▼
[Agent 2 — Decision AI]  gemma2:2b / kimi-k2.5:cloud
Filters sections by relevance to the query (YES/NO)
    │
    ▼
[Agent 3 — Reasoning AI] deepseek-r1:7b / glm-5:cloud
Generates final answer from filtered context only
    │
    ▼
Answer + Source Sections

---

Model Configuration

Query3AI uses a "Mix and Match" architecture, meaning you can configure different providers (Local, OpenAI, Groq, etc.) for each of the three agents simultaneously.

Edit ~/.query3ai/config.json to customize models, keys, and base URLs:

{
    "TREE_API_KEY": "sk-proj-...",
    "DECISION_API_KEY": "gsk-...",
    "REASONING_API_KEY": "",
    "TREE_MODEL": "openai/gpt-4o",
    "DECISION_MODEL": "groq/llama-3.3-70b-versatile",
    "REASONING_MODEL": "ollama/qwen3-32b",
    "TREE_API_BASE": "",
    "DECISION_API_BASE": "",
    "REASONING_API_BASE": "http://localhost:11434"
}

This allows you to leverage the best model for each specific task (e.g. OpenAI for high-context tree structuring, Groq for lightning-fast decision processing, and local Ollama for reasoning security).

---

Interactive Chat

Start the TUI chat interface:

query3ai chat

Slash Commands

Command	Action
/about	Learn about Query3AI
/help	Display all available commands
/ingest <path>	Ingest a new document
/listdocs	List all indexed documents
/list	Show total sections and chunks
/deletedoc	Remove a document from database
/cleanupdocs	Delete all documents
/cleanupresorce	Clean up temporary files
/clear	Clear terminal
/exit	Exit chat

---

Requirements

Component	Minimum	Recommended
RAM	8 GB	16 GB
CPU	4 cores	8 cores
GPU	Not required	Optional
Python	3.8+	3.10+
Storage	10 GB free	20 GB free

---

Tech Stack

Layer	Technology
CLI	Typer + Rich
AI Inference	Ollama, Groq
Local Models	phi3.5, gemma2:2b, deepseek-r1:7b
Cloud Models	qwen3.5:cloud, kimi-k2.5:cloud, glm-5:cloud
Graph Database	Neo4j
Document Parsing	PyMuPDF, python-docx

---

License

MIT