WebX

GhostOperator AI is a local-first AI agent that transforms natural-language instructions into structured, multi-step computer workflows. The project addresses the challenge of automating repetitive file-management tasks without requiring users to manually perform every operation or rely on unrestricted AI-generated terminal commands. Built using Python and Meta's Llama 3.2 model through Ollama, GhostOperator follows a modular architecture: - Workspace Context Engine: Scans the project workspace and gathers file metadata, including names, extensions, sizes, and modification times. - AI Planner: Uses Llama 3.2 to interpret the user's instruction and generate a structured JSON execution plan. - Safety Review: Checks whether the proposed plan uses supported tools and performs an additional local AI-based review. - Execution Engine: Executes predefined Python operations such as finding the latest CSV file, extracting rows matching a keyword, and saving the results. The user reviews and confirms the generated plan before execution. In my demonstration, the user asks the agent to find the latest CSV file, extract all rows containing "ERROR", and save them into an error report. GhostOperator generates the plan and executes the corresponding operations. Our goal is to explore how local language models can power practical AI agents while keeping execution controlled and users involved in important actions.