Stop burning tokens on solved problems: How my 12-year-old Office PC handles Email, OCR, and TTS for $0

Hi everyone, I’m running a small OpenClaw Telegram bot that helps me manage my To-Do lists and appointments. My server (Herbert) is an ancient office PC powered by an Intel Core i5-4460S—which is now 12 years old. At first, I was burning through API tokens like I was actively trying to get rid of them. Then I realized I was using expensive LLM calls for problems that software solved decades ago. Here is how I pushed most of my bot's routine tasks back to local, zero-token scripts (plus a little psychological productivity hack): 1. Email Checking & Pre-Processing The mistake: Asking an LLM to check a dedicated bot mailbox every 5 minutes. The fix: Let a simple Python script poll the inbox via IMAP. It filters spammers, strips out HTML metadata, and only hands the distilled plain-text content to the model if action is required. Zero tokens burned for idle polling. 2. Document & Image OCR (Tesseract) The mistake: Uploading screenshots, photos of leaflets, or PDFs directly to multimodal LLMs (which costs a fortune in vision tokens). The fix: I installed tesseract locally. Even this 12-year-old CPU extracts the text in about 3 seconds. The LLM only receives the clean OCR text output. It’s vastly cheaper and works like a charm. 3. Voice Messages (Piper TTS vs. OpenAI) The mistake: Paying OpenAI per character for TTS to send daily weather/calendar briefings as voice notes (and its German sounded pretty terrible with a heavy US accent). The fix: I switched to Piper TTS running locally via ONNX. It takes about a third to half the message's duration to render on my old CPU, but I easily have that time. Best of all: The local German voice models sound way better and more natural than OpenAI ever did. 4. The "TaDa-List" (Instead of just To-Dos) The fix: Beyond normal task management, my bot handles a "TaDa-List" in Todoist. Instead of just tracking what I failed to do, it logs the unexpected, chaotic things I actually accomplished throughout the day that weren't even on any schedule. It’s a zero-token psychological boost that keeps me sane when the actual To-Do list feels overwhelming. What basic or routine tasks in your agent setups have you realized you could easily push down to local, zero-token scripts (or what non-tech hacks do you run)?