CmdSmith Project Log

Goal: Build a "Fill-in-the-Blank" CLI generator for researchers using RAG (Documentation) and LLMs (Structured Outputs). Stack: Python (FastAPI), LangChain (RAG), ChromaDB (Vector Store), OpenAI GPT-5.1/Local LLM, HTML/JS (Frontend).

---

1. Initial Setup

A. Project Folder & Virtual Environment

Never install libraries globally. We use a contained venv.

# 1. Create virtual environment
python3 -m venv venv

# 2. Activate the environment
source venv/bin/activate

B. Dependencies

Install them:

pip install -r requirements.txt

C. API Keys

Create a file named .env in the project root:

OPENAI_API_KEY=sk-api-key

---

2. Code Implementation

A. The Ingestion Script (ingest.py)

Purpose: Reads PDF/TXT manuals and saves them to ./vector_db.

B. The Backend (main.py)

Purpose: API that takes user input, retrieves docs, and returns structured JSON. Key Feature: Uses specialized system prompt and Strict JSON Schema.

C. The Frontend (index.html)

• Use the "Dark Mode" theme.

---

3. Workflow: How to Update Documentation

1. Find Manual: Save the tool's manual as a PDF (e.g., samtools.pdf).
2. Place: Put it in the docs/ folder.
3. Alternatively, scrape command descriptions using scrape.py.
4. Run Ingest:

   source venv/bin/activate
   python ingest.py
   

   This updates the vector_db folder.

---

4. Usage: Running the Tool

1. Open Terminal and go to project folder.
2. Activate Environment:

   source venv/bin/activate
   

3. Start Backend:

   python main.py
   

   (Wait for "Application startup complete")
4. Start Frontend:
   • Open index.html in web browser (Chrome/Safari).
5. Use:
   • Start chatting

---