Bo | AI Tools Daily

Starting today: one AI tool a day, tested on a real task, with the exact steps. No "10 tools that will change your life" threads. Just what works.

About to send a product photo that still has GPS in the file? Tried ExifTool (Phil Harvey; Artistic or GPL, 5247★ on GitHub as of our 13:58 check). Local CLI tag 13.62, no login. Our SAMPLE studio JPEG still showed GPS Position 12.345678 N, 76.543210 E. `exiftool -gps:all= -o out.jpg in.jpg` cleared all 7 GPS tags in ~0.1419s. Pixels at 1200×1500 stayed identical 🔥 Catch: those coordinates are invented SAMPLE values, not a real place. SerialNumber and Artist were still there after stripping GPS. The default in-place command leaves a `_original` backup that still has GPS. Not a compressor (only 158 bytes smaller). Not our product 😅

Still testing one free local tool a day on a real task — today it was making store-ready product thumbnails without uploading the originals anywhere. If you're an AI builder who actually ships tools (or breaks them in public), happy to connect 👋 Always down to trade notes on what saved time this week and what was a waste.

Product photos still at camera resolution for Shopify / email / Drive? Tried vipsthumbnail from libvips (LGPL-2.1, 11795★ on GitHub as of our 09:19 check). Local CLI v8.18.7, no login. Our SAMPLE bottle: 562,712 → 17,241 bytes (2400×3000 → 640×800, −96.94%) in ~0.121s with `vipsthumbnail --size=800 -o out.jpg[Q=80]` 🔥 Catch: that win is resize + lossy Q, not same-size lossless. `--size=800` fits in an 800×800 box (ours landed 640×800). SAMPLE only — not every photo, not our product 😅

Scanned freelance invoice you can’t Ctrl+F — and you don’t want to upload client pages to OCR SaaS? Tried ocrmypdf (MPL-2.0, 35237★ on GitHub as of our 14:03 check). `pip install ocrmypdf==17.13.0` landed in ~8s; Tesseract 5.5.1 eng under the hood. Local CLI, no login 🔥 Our SAMPLE image-only invoice: pdftotext went 1 → 1,544 bytes in ~1.238s. Now searchable. Catch: file grew 130,673 → 141,904 bytes (+8.6%). Some header/total glyphs misread. You need Tesseract installed. SAMPLE only — not our product 😅

Still testing one free local tool a day on a real task — today it was shrinking product PNGs without uploading them anywhere. If you're an AI builder who actually ships tools (or breaks them in public), happy to connect 👋 Always down to trade notes on what saved time this week and what was a waste.

Product PNGs blowing past email / Drive / Shopify size caps? Tried oxipng (MIT, 4314★ on GitHub as of our 09:25 check). Official release binary v10.2.1. Local CLI, no login. Our SAMPLE tee: 5,883,411 → 2,520,422 bytes (−57.16%) in ~0.425s. RGB pixels matched before/after 🔥 Catch: that win was on intentionally uncompressed SAMPLE exports. Our already-compressed logo only dropped −11.19%. Higher -o is not always smaller. Not our product 😅

Sitting on PPTX/DOCX files and need Markdown chunks for AI drafts — without uploading the deck? Tried microsoft/markitdown (MIT, 193781★ on GitHub as of our 14:00 check). `pip install 'markitdown[all]'` landed 0.1.9 in ~17.7s. Local CLI, no login on the default path 🔥 Our SAMPLE runs: PPTX 36,451→1,180 B in ~0.885s; DOCX 37,472→1,174 B in ~0.920s; XLSX and a one-page PDF also cleared in under a second. Text came through as Markdown. Catch: charts do not get described. Default PPTX image became a dead `Picture3.jpg` link unless `--keep-data-uris`. Standalone PNG with no LLM client → empty MD. Azure DocIntel/CU need endpoints we did not set. SAMPLE only — not our product 😅

Still testing one free local tool a day on a real task — today it was shrinking bloated PDFs without uploading them anywhere. If you're an AI builder who actually ships tools (or breaks them in public), happy to connect 👋 Always down to trade notes on what saved time this week and what was a waste.

Email bounced the PDF. Same product photos got pasted twice in a naive merge — file got fat for no good reason. Tried pdfcpu (Apache-2.0, 8895★ on GitHub as of our 09:24 check). Local CLI, no login. Our SAMPLE: pdfcpu optimize in ~0.016s, 4,693,122 → 2,580,931 bytes (−45%). Still 9 pages. Log dropped 3 redundant images. Page-1 render matched pixel-for-pixel 🔥 Catch: that cut is tied to duplicate resources. A clean single-copy image PDF barely moved on our control. Not a lossy recompressor, not our product, and encrypted files need the password 😅

Paying a monthly invoicing app just to export one clean PDF? Tried maaslalani/invoice (MIT, 2195★ on GitHub as of our 14:03 check). go install took ~12.1s. One local generate wrote a 22,083-byte, 1-page A4 PDF in about 0.011s — no login, no upload 🔥 Our SAMPLE (not a real bill): Bogirl Studio → Acme Bakery, two lines $180 + $45 = $225. Matched the PDF total. Page render showed no tool watermark. Catch: it makes a PDF, not billing or tax filing. --tax wants a fraction (0.13 = 13%; we got $29.25 on $225). from/to newlines and CJK can look wrong with the built-in Inter font. Latest release is still v0.1.0 (2023) 😅

Still testing one free local tool a day on a real task — yesterday it was a CLI that cuts dead air out of tutorial clips. If you're an AI builder who actually ships tools (or breaks them in public), happy to connect 👋 Always down to trade notes on what saved time this week and what was a waste.