What do you mean by “agent”? Not an autonomous bot. It’s a constrained system that operates over spreadsheet primitives (tables, ranges, dependencies) with explicit actions.
How is this different from Copilot / formula generation? Most tools generate formulas or text and stop there. This operates on sheet structure directly — extraction, normalization, multi-sheet modeling — and leaves behind editable, inspectable spreadsheets rather than opaque outputs.
Why not Excel + Python? That works well for some workflows, but many users want everything to stay inside the spreadsheet execution model where recalculation, auditing, and collaboration already live.
https://github.com/raj-khare/offset
Hi HN — working with spreadsheets still requires a lot of manual glue work.Common pain points we kept running into: • Copy-pasting data from PDFs and reports into sheets • Cleaning and formatting inconsistent exports before any real work can start • Building multi-sheet financial models that are fragile and hard to change • Manually wiring spreadsheets to external data sourcesMost AI...
Hey HN, I built SideSpark because I was frustrated with cloud note-taking tools and their subscription creep.SideSpark runs locally on macOS using on-device models. Nothing leaves the machine. No data collection. No recurring fees. Just a simple, private note taker that works offline.Would love feedback, critiques, and ideas for improvement.
Hey HN,I built SideSpark because I was frustrated with cloud note-taking tools and their subscription creep.SideSpark runs locally on macOS using on-device models. Nothing leaves the machine. No data collection. No recurring fees. Just a simple, private note taker that works offline.Would love feedback, critiques, and ideas for improvement.
We just launched recently (https://x.com/ofstapp/status/1972741248495833219)—an autonomous financial analyst that handles the stuff everyone dreads in Excel or Sheets: formatting, financial modeling, data enrichment. Work that usually eats up hours of grunt time can now be done in minutes.
Try it out today!
Happy to answer any questions :)
It seems to me that this is a new ground up architecture designed to run deep learning algorithms with orders of magnitude less energy than current GPUs, TPUs. I know current deep learning leaders are attacking this field (airplanes don't flap wings), but I think if people work on it, it could be a great deal. (It was a stupid idea to work on NNs when SVM were better in 90s.)
I'm not saying I'm not against people who are abusing H1B. That should be banned NOW.