An advanced CLI terminal-based AI assistant designed to extract, analyze, and discuss PDF documents with high technical rigor. This tool is built for professionals who require deep document intelligence without compromising data privacy.
The primary function of this assistant is to serve as a Senior Expert Panel. Unlike cloud-based solutions, this tool is fully optimized to run locally on your machine.
- 100% Local Execution: By leveraging Ollama, all data processing and AI inference happen on your hardware.
- Security & Confidentiality: Your sensitive documents never leave your local environment. There are no external API calls to third-party providers, ensuring total data sovereignty.
- CLI Optimized: Purpose-built for speed and efficiency via the command line.
Note: While this version is optimized for the Command Line Interface (CLI), a web-browser version is currently under development to provide a graphical interface in the future.
To ensure maximum performance, the script implements a Pickle-based caching system.
- High-Speed Reloading: The first time a PDF is analyzed, the script performs a deep extraction and sanitizes the content.
- Persistence: It saves the processed data into a
.pklfile. Subsequent runs load this data instantly, bypassing the extraction phase for immediate interaction.
The tool employs a sophisticated extraction logic to handle complex document layouts:
- Table Preservation: Uses spatial block-sorting to maintain the logical flow of rows and columns, allowing the AI to interpret data structures and financial tables accurately.
- Object Filtering: The engine explicitly ignores images and non-text graphics. This reduces noise, saves context tokens, and focuses the AI purely on the textual and tabular intelligence of the file.
The assistant organizes its findings into four distinct expert phases, each designed to provide a specific layer of document intelligence:
- PHASE 1: CLASSIFICATION – Strategic identification of the document type, domain, and intent.
- PHASE 2: HIGHLIGHTS – A comprehensive summary of key themes and structural pillars.
- PHASE 3: TOP 10 CRITICAL INSIGHTS – Extraction of the most vital technical, legal, or financial points.
- PHASE 4: EXECUTIVE SUMMARY – A high-level final verdict for decision-makers.
Interactive Expert Loop: Following the initial report, the tool opens a continuous interactive session. It maintains full context memory of the document.
This tool requires Ollama to run the LLM locally.
- Download Ollama from ollama.com.
- Install the Llama 3 model by running:
ollama pull llama3
The project is configured with a pixi.toml file for reproducible environments.
- Install Pixi from pixi.sh.
- To set up the environment:
pixi install
The project includes a Pixi Task alias for faster execution. You can use the analyze shortcut:
pixi run analyze --pdf <path_to_pdf> --detail <1-10>
(This alias maps to: python app.py as defined in pixi.toml)
Parameters:
--pdf: (Required) Path to the PDF file you want to analyze.--detail: (Optional) Level of depth for the report, from 1 to 10 (Default: 5).
Interactive Session Commands:
- Ask anything: Just type your question about the document and press Enter.
- clear: Resets the chat history memory.
- exit / quit: Safely closes the session.
Output:
- The initial report is automatically saved to a file named
summary.mdin the root directory.