Jarvis is an autonomous agent designed specifically for Developer Workflows. Unlike generic chatbots, Jarvis is an active pair programmer that lives in your environment, debugging code, managing files, and automating repetitive engineering tasks with human-in-the-loop safety.
- Problem: Most agents crash when their generated code fails.
- Solution: Jarvis captures
stderr, feeds it back to the LLM, analyzes the root cause, and rewrites the script automatically. It iterates until the code works or a maximum retry limit is reached. - Use Case: "Fix this script" -> Jarvis runs it, sees the error, patches it, and verifies the fix.
- Feature: Smart Window Analysis.
- How it works: Jarvis detects your active window (e.g., "main.py - VS Code") to infer context. If you say "Debug this", it knows exactly what "this" refers to without needing a file upload.
- Philosophy: AI should be powerful but controllable.
- Implementation:
- Read-Only Default: Jarvis can read screens and files freely.
- Active Confirmation: Any destructive action (File Deletion, Shell Execution, System Shutdown) requires explicit user approval via a native OS dialog.
- Visual Indicator: The UI clearly signals when Safe Mode is active (Green Shield).
- Feature: Real-time access to documentation and tech stacks.
- Tech: Integrates Tavily/Serper APIs to fetch the latest API references, library updates, or GitHub issues, ensuring code suggestions aren't outdated.
- Python 3.9+
- Google Cloud API Key (Gemini) OR Groq API Key
-
Clone the repository:
git clone https://github.com/your-repo/jarvis-ai.git cd jarvis-ai -
Install Dependencies:
pip install -r requirements.txt
-
Configure Environment:
- Rename
.env.exampleto.envand fill in your keys:# AI Provider (gemini or groq) AI_PROVIDER=gemini # Keys GOOGLE_API_KEY=your_gemini_key GROQ_API_KEY=your_groq_key TAVILY_API_KEY=your_tavily_key_here # Optional for Deep Search
- Rename
-
Run:
python main.py
- Current State (MVP): Jarvis operates within a defined
workspace/directory for file creation to prevent accidental system clutter. - Production Roadmap: The production version will utilize ephemeral Docker Containers for each task execution to ensure complete isolation and security.
gui.py: Modern PyQt5 interface with non-intrusive notifications and status cycling.ai_engine.py: Manages Gemini 2.0 Flash sessions with a specialized "Developer Persona" system prompt.task_manager.py: Handles tool execution with robust error handling and Safe Mode enforcement.workflow_learner.py: Experimental module for learning repetitive GUI tasks via computer vision.
Run the unit test suite to verify core functionality:
python -m unittest discover testsBuilt for Hackathon 2026. Focused on practical developer utility, not hype.