Launched this week
Spredly
Talk to your spreadsheets directly, use local LLMs or Claude
10 followers
Talk to your spreadsheets directly, use local LLMs or Claude
10 followers
Spredly is a modern spreadsheet experience built for people who want the speed of AI without giving up the familiarity of rows, columns, formulas, and exports. You can import data, chat with your spreadsheet in plain English, generate charts, and turn raw tables into useful dashboards in minutes. Unlike generic AI assistants that sit outside your workflow, Spredly works directly inside the spreadsheet experience. Upd 3.05.26 Added "find inputs" feature with cell type highlighting.
The local LLM angle is exactly right for finance work — anyone who's built real deal models knows you can't be feeding sensitive deal data into cloud APIs without a proper review process. Local processing removes that friction entirely.
From a financial modeling perspective, I'd love to see support for the structured layout patterns deal teams rely on — color-coded assumption cells, hardcoded vs. formula distinctions, named range navigation across a 50-tab model. Been working in this space with ModeLoop and the pain point is very real. The natural language query layer here could be a genuine time-saver on large model reviews. Congrats on shipping!
@samir_asadov Thank you 😊 🙏 I'll try to include these features you mentioned in the upcoming versions of the app 🫶
@samir_asadov By the way when I implement features that you've mentioned, may I send you the offline version of the app for testing (in exchange for your feedback) ? Im really glad that you can test it for something immidiately useful and valuable to you.
If you get the local app for Mac/Windows you can plug your agent via localhost:4141. All spreadsheet related actions will be available to your agent and you will see everything that it does live inside the app.
RiteKit Company Logo API
@vitalyiam That's a smart workaround for the premium paywall problem. Local LLMs give you way more control and privacy with sensitive data too—curious if you've noticed any performance differences compared to the cloud-based options when working with larger datasets.
@osakasaul Thank you 🙏! Yes, actually local models around 9b and up are pretty fast with any type of prompt (maybe 20s per complex chart max). I've made it so the agent gets only the data it needs to make changes, which made all interactions quicker
Added "find inputs" feature.
Find Inputs scans the whole workbook and highlights the cells that matter most for review: formulas, hardcoded numbers, styled input cells, and likely assumptions. It works across all sheets, preserves Excel formula metadata on import, and lets users quickly spot which values are calculated versus manually entered without sending sensitive data to the cloud.
API is not required for this feature to work.