pydebugai 0.1.0

AI-powered Python debugging assistant — ChatGPT for Python errors

PyDebugAI 🐛

AI-powered Python debugging assistant — like ChatGPT, but exclusively for Python errors. Installable via pip. Runs in your terminal and in VSCode.

---

Features

Feature	Description
🔍 AST Static Analysis	Catches errors before running — undefined names, division by zero, type mismatches
🤖 Rule Engine	60+ hand-crafted rules for instant, deterministic fixes
🧠 ML Classifier	TF-IDF + Random Forest classifies error types with high confidence
📝 Levenshtein Fixer	Detects typos in variable/function names and suggests corrections
🚀 CodeBERT	Deep semantic code understanding via HuggingFace transformers
📚 Self-Learning	SQLite + RLHF-lite — improves from every upvote/downvote
🧩 VSCode Extension	Red squiggles, hover hints, quick-fix menu, sidebar chat panel
💻 CLI Runner	pydebugai run file.py — run and debug from your terminal

---

Quick Start

1. Install via pip

pip install pydebugai

Or install from source (this repo):

cd c:\new-model-ai
pip install -e .

2. Run the CLI

# Analyze a Python file
pydebugai run my_script.py

# Interactive chat mode (paste code or errors)
pydebugai chat

# Start the local API server (required for VSCode extension)
pydebugai serve

# Show self-learning statistics
pydebugai stats

3. Install the VSCode Extension

# Install from .vsix (after packaging)
code --install-extension vscode-extension/pydebugai-0.1.0.vsix

Or in VSCode: Extensions → Install from VSIX...

Then start the server in your terminal:

pydebugai serve

The extension will:

• Show red squiggles on error lines
• Display hover hints with fix suggestions
• Offer 💡 Fix with PyDebugAI in the quick-fix menu
• Open the sidebar panel with full AI analysis

---

AI Algorithms

User Code
    │
    ├─► AST Analyzer (static, instant)
    │       └─ SyntaxError, IndentationError, undefined names,
    │          division by zero, comparison style warnings
    │
    ├─► Code Executor (subprocess, safe)
    │       └─ Captures stdout / stderr / exit code
    │
    ├─► Error Parser (regex traceback parser)
    │       └─ Extracts: exception type, message, file, line, col
    │
    ├─► Rule Engine (60+ rules, deterministic)
    │       └─ Pattern-match → instant fix suggestion
    │
    ├─► ML Classifier (TF-IDF + Random Forest)
    │       └─ Classifies error type → boosts rule confidence
    │
    ├─► Levenshtein Fixer (edit distance)
    │       └─ "Did you mean 'print'?" for NameError/AttributeError
    │
    ├─► CodeBERT (HuggingFace, optional)
    │       └─ Semantic embeddings → cosine similarity retrieval
    │
    └─► Orchestrator → Merge, Deduplicate, Rank by confidence
                └─► Top-10 suggestions → CLI / VSCode / API

Self-Learning Loop (RLHF-lite)

User gives thumbs-up / thumbs-down on suggestion
    → Stored in SQLite (pydebugai/data/feedback.db)
    → Positive samples accumulate
    → Every 25 positive samples → auto-retrain ML classifier
    → Model improves over time per user

---

CLI Examples

# Run and auto-debug a buggy file
pydebugai run tests/samples/buggy_sample.py

# Get output as JSON (pipe to other tools)
pydebugai run my_script.py --json

# Static analysis only (don't execute)
pydebugai run my_script.py --no-exec

# Deep analysis with CodeBERT (slower)
pydebugai run my_script.py --deep

# Interactive REPL
pydebugai chat
>>> x = 10 / 0
# (press Enter twice)
# → AI prints fix suggestions

---

API (for VSCode extension & custom integrations)

Start server: pydebugai serve (default: http://localhost:7432)

# Analyze code
curl -X POST http://localhost:7432/analyze \
  -H "Content-Type: application/json" \
  -d '{"code": "print(udefined)"}'

# Record feedback
curl -X POST http://localhost:7432/feedback \
  -H "Content-Type: application/json" \
  -d '{"interaction_id": 1, "selected_idx": 0, "feedback": 1}'

# Health check
curl http://localhost:7432/status

# Stats
curl http://localhost:7432/stats

---

Requirements

Minimal (fast startup, no ML)

click, rich, fastapi, uvicorn, pydantic

Full AI features

scikit-learn    # ML classifier
transformers    # CodeBERT
torch           # PyTorch backend
python-Levenshtein  # fast edit distance

Install all:

pip install -r requirements.txt

---

Running Tests

cd c:\new-model-ai
python tests/test_error_parser.py
python tests/test_ast_analyzer.py
python tests/test_rule_engine.py

---

VSCode Extension Settings

Setting	Default	Description
pydebugai.serverPort	7432	Local server port
pydebugai.analyzeOnSave	true	Auto-analyze on save
pydebugai.enableExecution	false	Run code during analysis
pydebugai.pythonPath	python	Python interpreter path

---

License

MIT © PyDebugAI Team