Robotic Process Automation (RPA)

Robotic Process Automation (RPA) is software that automates repetitive, rule-based tasks typically performed by humans interacting with graphical user interfaces (GUIs). RPA bots mimic mouse clicks, keystrokes, and data entry across applications like spreadsheets, email clients, and enterprise systems. Unlike AI agents that make decisions or reason, RPA follows predefined scripts and triggers. Operators encounter RPA as a complementary tool to local AI: an RPA bot can feed data into a local LLM for processing, then act on the LLM's output—for example, extracting text from a PDF, sending it to a local model for summarization, and pasting the summary into a CRM.

RPA operates at the UI layer, often using screen scraping, OCR, or API calls to interact with applications. It is typically deployed via platforms like UiPath, Automation Anywhere, or open-source alternatives like TagUI. RPA is not AI—it lacks learning or adaptation—but it can be combined with local AI to handle unstructured data. For example, an RPA bot might collect customer emails, pass them to a local LLM for sentiment analysis, and then route them to the appropriate department. The key distinction: RPA automates the 'how' (clicking, copying, pasting), while AI automates the 'what' (understanding, generating). On a local rig, RPA scripts run on the CPU and consume minimal resources, making them lightweight companions to GPU-bound AI workloads.

An operator runs an RPA bot that monitors a shared folder for incoming invoices (PDFs). The bot extracts text using OCR, then calls a local LLM via Ollama's API (e.g., curl http://localhost:11434/api/generate -d '{"model":"llama3.2","prompt":"Summarize this invoice: ..."}'). The bot parses the LLM's JSON response, extracts the total amount and due date, and enters them into a spreadsheet. This workflow runs on a single machine with 16 GB RAM—the RPA bot uses <100 MB, leaving most resources for the LLM.

In a typical setup, an operator uses an RPA tool like UiPath or a Python script with pyautogui to automate data entry. The RPA script launches a browser, logs into a web portal, and scrapes a table. It then sends each row to a local LLM (via LM Studio's local API) for classification. The LLM returns a label, and the RPA bot updates the portal accordingly. The operator debugs by checking the RPA log for UI element failures and the LLM log for response quality. This hybrid approach offloads decision-making to the LLM while RPA handles the repetitive clicking.