Drive-Claw: Google Drive Discovery Terminal

Drive-Claw is a conversational AI agent engineered to bridge the gap between human intent and the structured chaos of your google drive. Built as a specialized "librarian" for Google Drive, it is an upgrade from simple keyword matching by translating natural language into high-precision API queries.

The Tech Stack

The architecture is built on a decoupled, production-ready stack designed for performance, modularity, and elegance:

• Backend: Python with FastAPI for high-performance asynchronous API routing.
• Agent Intelligence: LangGraph (Cyclic State Machine) for advanced orchestration, state management, and recursive reasoning.
• LLM Engine: Powered by gemma-4-26b (via langchain-google-genai) for rapid intent recognition and tool calling.
• Frontend: Streamlit crafted with a beautiful easy on the eyes and bespoke UI/UX. (I love designing cool UIs)
• Integration: Google Drive API (v3) utilizing Service Accounts for secure, autonomous access.
• Containerization: Docker for seamless, environment-agnostic deployment.
• Hosting: Hosted on Render free-tier cloud environments.

The Problem I Aimed Solved

Traditional file searching often requires exact keyword matches or tedious manual navigation through nested folders. Drive-Claw solves the "Discovery Friction" by:

• Intent Translation: Understanding what a user wants (e.g., "Find the financial report from last week") and translating it into a valid Google Drive q parameter.
• Deep Discovery: Bypassing the recursive limitations of the Drive API to peer into the deepest subfolders of a repository.
• Contextual Awareness: Maintaining a "Memory" of the conversation to allow for natural, back-and-forth clarification and exploration.

Key Features

• Natural Language Discovery: Support for exact/partial name matches, MIME type filtering (PDFs, Docs, Sheets, Images), and full-text content searching.
• Recursive Deep Search: Custom logic that allows the agent to find nested files two or more directories deep.
• Temporal Reasoning: Real-time calculation of relative dates (e.g., "yesterday," "last month") into RFC 3339 timestamps.
• Clinical UI/UX: A sleek interface with custom CSS, visual "thinking" cues (spinners), and responsive chat bubbles.
• Security First: Strictly utilizes Service Account authentication and Environment Variable injection (no hardcoded credentials).
• Tool Calling Logic: Dedicated DriveSearchTool that prevents LLM hallucinations by enforcing strict query syntax.

Build on your own

Prerequisites

• Python 3.12+
• Google Cloud Project with Drive API enabled.
• Service Account JSON key and a Gemini API Key.

Local Installation

1. Clone the Repo: git clone https://github.com/type-abhay/drive-speak/
2. Environment Setup: Create .env files in both backend/ and frontend/ folders with your API keys and folder IDs.
3. Docker Build:

   # From root
   docker-compose up --build

   Alternatively, run the backend and frontend separately using uvicorn and streamlit.

Deployment Note

This project is configured for deployment on Render using the provided Dockerfiles. Ensure that GOOGLE_CREDENTIALS_JSON and BACKEND_URL are correctly set in your cloud environment variables to facilitate the "Synaptic Handshake" between services.

---

Thank You

Thank you for exploring Drive-Claw. This project was a journey into the physics of materials—translating messy, organic human language into the rigid, logical structures of machine query. It stands as a testament to the power of modern agentic frameworks and clinical UI design.

Stay curious, stay sharp.