https://github.com/alexg-4w/autonomous-engineering-copilot
The Autonomous Engineering Copilot is a cross-platform Python desktop application designed to act as an offline AI assistant. By utilizing local LLMs and embedded vector databases, it provides powerful RAG capabilities entirely on the user's hardware.
https://github.com/alexg-4w/autonomous-engineering-copilot
ai-assistant lancedb llama-cpp local-llm offline pyside6 rag
Last synced: about 2 months ago
JSON representation
The Autonomous Engineering Copilot is a cross-platform Python desktop application designed to act as an offline AI assistant. By utilizing local LLMs and embedded vector databases, it provides powerful RAG capabilities entirely on the user's hardware.
- Host: GitHub
- URL: https://github.com/alexg-4w/autonomous-engineering-copilot
- Owner: AlexG-4W
- License: agpl-3.0
- Created: 2026-03-16T02:46:32.000Z (2 months ago)
- Default Branch: main
- Last Pushed: 2026-03-31T23:52:45.000Z (about 2 months ago)
- Last Synced: 2026-04-01T01:54:36.913Z (about 2 months ago)
- Topics: ai-assistant, lancedb, llama-cpp, local-llm, offline, pyside6, rag
- Language: Python
- Homepage: https://alexg-4w.github.io/
- Size: 380 KB
- Stars: 0
- Watchers: 0
- Forks: 0
- Open Issues: 0
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
README
# Autonomous Engineering Copilot 🤖
## Vision
The Autonomous Engineering Copilot is a cross-platform desktop application designed to act as an offline AI assistant. By utilizing local Large Language Models (LLMs) and embedded vector databases, it provides powerful RAG (Retrieval-Augmented Generation) capabilities entirely on the user's hardware.
## ✨ Key Features & Use Cases
- **Local Document Q&A:** Users can seamlessly query proprietary technical specifications, datasheets, and compliance documents directly on their machine, receiving accurate insights with specific page references.
- **Multiple File Formats:** Support for `.pdf`, `.csv`, `.docx`, `.xlsx`, `.txt`, and `.md` file ingestion.
- **Multi-Table Vector Search:** Dynamically isolates each uploaded document into its own LanceDB table, allowing users to query "All" documents via Reciprocal Rank Fusion, or filter the search to a specific document collection via the UI sidebar.
- **Source Citations:** Automatically appends document sources and metadata to the end of the AI's response so users can verify facts.
- **Modern UI & Configurable Inference:** A sleek dark theme (Fusion) with a GUI settings panel for dynamically adjusting LLM generation parameters (e.g., Temperature, Top-P, Model Selection, Context Chunks).
- **Real-Time Streaming:** The AI copilot streams generation text incrementally into the chat window for immediate feedback and responsiveness.
- **Infrastructure Independence:** The system operates without any requirement for API keys, internet connectivity, or specialized cloud infrastructure, ensuring uninterrupted access and complete data privacy.
# 🚀 Release v2.1-alpha: Gemma 4 Support
## 🔥 Highlight: Full Gemma 4 Compatibility
The core inference engine has been upgraded to support the newly released **Gemma 4** architecture. The application now seamlessly handles the advanced tokenization and multi-layered embeddings of the latest GGUF models.
* **Recommended Model:** This release is optimized and tested with the `Gemma-4-E4B-Uncensored-HauhauCS-Aggressive-Q4_K_M.gguf` model for precise, unfiltered technical generation.
* **Download the Model:** [Download Gemma-4-E4B here](https://huggingface.co/HauhauCS/Gemma-4-E4B-Uncensored-HauhauCS-Aggressive#aggressive-variant)
## 🛠️ Quick Start Instructions
1. Download the `AECopilot2.1-alpha.zip` archive from the assets below and extract it to a folder on your PC.
2. Download your preferred `.gguf` model (like the recommended Gemma 4 model above).
3. Place the downloaded `.gguf` file inside the `AECopilot/model/` directory.
4. Run `AECopilot.exe` and start querying your local documents entirely offline!
---
## 🛠️ Building & Running from Source
If you want to modify the code, run it via Python, or build your own executable, follow these steps.
### 1. Prerequisites
- Python 3.11+
- Git
### 2. Setup Virtual Environment
```bash
python -m venv .venv
# On Windows:
.venv\Scripts\activate
# On Linux/macOS:
source .venv/bin/activate
```
### 3. Install Dependencies
```bash
pip install -r requirements.txt
```
## Troubleshooting
### Failed building wheel for llama-cpp-python (Windows)
This error occurs during dependency installation if your system lacks the necessary C++ compiler to build the `llama-cpp-python` package from source.
**Solution:**
1. Download [Visual Studio Build Tools](https://visualstudio.microsoft.com/visual-cpp-build-tools/) from Microsoft.
2. Run the installer and select the **"Desktop development with C++"** workload.
3. Finish the installation and **restart your terminal or IDE** (e.g., VS Code) to update the system paths.
4. Re-activate your virtual environment and run `pip install -r requirements.txt` again.
### ModuleNotFoundError: No module named 'PySide6'
This usually happens if the `pip install -r requirements.txt` command failed or was interrupted (often by the C++ compiler issue mentioned above). Because the process stopped early, subsequent packages like `PySide6` were not installed.
**Solution:** Resolve any build errors first, then re-run `pip install -r requirements.txt` to ensure all required packages are successfully installed into your `.venv`.
### 4. Download Model
Download a GGUF model (e.g., [Llama-3-8B-Instruct-GGUF](https://huggingface.co/NoelJacob/Meta-Llama-3-8B-Instruct-Q4_K_M-GGUF) , [Qwen2.5-7B-Instruct-Q4_K_M.gguf](https://huggingface.co/WSDW/Qwen2.5-7B-Instruct-Uncensored-Q4_K_M-GGUF) , [Mistral-Nemo-Instruct-2407-Q4_K_M.gguf](https://huggingface.co/Nehal07/Mistral-Nemo-Instruct-2407-Q4_K_M-GGUF) ) and place it in the `model/` directory.
```bash
mkdir model
# Place your .gguf file here
```
### 5. Run the Application
You can run the application by double-clicking the `run_gui.bat` script (on Windows), or by running the module from your terminal:
```bash
python -m src.gui.main_window
```
---
## 📦 Packaging the Application (PyInstaller)
This project includes a custom automated build pipeline that compiles the PySide6 app and handles heavy Machine Learning libraries (like `torch`, `sentence-transformers`, and `lancedb`) correctly.
To build the executable yourself:
```bash
python build.py
```
**What the build script does:**
- Cleans up old cache and build artifacts.
- Uses **PyInstaller** to compile the app in `--onedir` (directory bundle) and `--noconsole` mode.
- Injects necessary metadata and hidden imports for PyTorch and Hugging Face architectures.
- Automatically copies the dynamic DLLs (`llama_cpp/lib`), the `model/` directory, and the `lancedb_data/` directory to the final `dist/AECopilot/` folder.
Once finished, your fully portable application will be waiting in the `dist/AECopilot/` directory!
---
## 🧪 Running Tests
To run the automated test suite, ensure your virtual environment is active and execute:
```bash
pytest tests/
```
## 📂 Architecture & Structure
- `src/gui/`: PySide6 frontend logic, asynchronous QThread workers.
- `src/ingestion/`: Extractors for PDFs (PyMuPDF) and CSVs.
- `src/rag/`: Logic for vector embedding, database interaction (LanceDB), LLM generation, and text chunking.
- `tests/`: Comprehensive test suite using pytest and MagicMock for GUI testing.
- `build.py`: Automated compilation pipeline for PyInstaller.