Show HN: I made a free (open-source) extension, to use any LLM on Google sheet
aisheeter.comHello everyone,
This past Saturday, I decided to make purchase on Cursor after hearing how good it is with Composer function that allow to generate code on fly, especially on Claude Sonnet (not an affiliate of the products, just curious!).
To make the purchase worthwhile, I challenged myself to build a Google Sheets extension in just 24 hours that allows users to connect with any Language Model (LLM)—whether it’s ChatGPT, Claude, Groq, or Gemini.
The Idea The idea was simple: a formula like =Claude(prompt, model) that lets users integrate AI anywhere in their sheets. The most impressive part? 80% of the process was just tabbing and prompting AI, while only *20% was actual coding The Tech Stack Frontend: React with Next.js Backend: Node.js with Next.js API routes Database: Supabase AI Integration: OpenAI API, Anthropic API, Groq API, and Google AI API Development Assistance: Cursor and Claude Sonnet 3-5
The Journey Planning (2 hours): I started by using Claude Sonnet 3-5 to brainstorm and outline the project. It handled so much of the heavy lifting when it came to feature ideas and technical suggestions.
Frontend Development (8 hours):With Cursor, I was literally tabbing through suggestions, and it felt more like _prompting_ AI to get the right code rather than writing it myself.
Backend Development (6 hours): The same pattern applied here: I spent most of the time *tabbing through Cursor’s suggestions and tweaking AI prompts* for the backend, which helped me set up secure API endpoints and efficient database schemas.
AI Integration (4 hours): This was the core part. I integrated multiple AI models to give users options. Claude helped refine the prompts and handle API responses with ease.
Testing and Debugging (3 hours): Cursor and Claude once again made it feel like I was navigating through the process rather than writing tests from scratch. Their combined tools flagged issues quickly, cutting debugging time in half.
Documentation and Deployment (1 hour): Drafted clear documentation with Claude’s help, and Cursor’s deployment features made production a breeze.
Key Features AI in Google Sheets: Use AI directly with simple formulas. Multiple AI Models: Support for ChatGPT, Claude, Groq, Gemini, and more. - Cost Tracking: Automatic token calculation to track usage costs. - Security: Encrypted API key storage for peace of mind.
Demo of the app : https://www.youtube.com/watch?v=HczW0F_tmvo
Challenges & Solutions The biggest challenge was integrating multiple AI models seamlessly. But with *Claude Sonnet 3-5* guiding the architecture and *Cursor* suggesting solutions, the integration was mostly about *tabbing and prompting*. The architecture now easily accommodates different AI services.
The extension is completely free at. : https://workspace.google.com/marketplace/app/aisheeter_smart...
What’s Next I named the project AISheeter.com - Sheets Smarter not Harder
I’m releasing it for free and open-sourcing the code [here](https://github.com/Ai-Quill/ai-sheet-frontend.git).
This is just the start—let’s see where it goes from here! Website is at : https://www.aisheeter.com/
Takeaway What stood out the most throughout this project was how much *AI could take over the coding process*. It was 80% navigating through suggestions and refining prompts, and only 20% traditional coding. With the right tools and mindset, it feels like the boundaries of what’s possible in such a short time are expanding.
Excited time ahead with AI !
This exists in CeLLM and various other implementations. Well done on a smooth version. The sad thing is that when one clicks in an LLM populated cell, one cannot see/edit the content as if it were normal text. One only gets the formula. So sad. Anyone think of a workaround ? Utility would be 10x otherwise.
Amazing work! The AI integration into Google Sheets with a simple formula is a game-changer, and completing it in 24 hours is impressive. The fact that 80% of the coding was just prompting AI highlights how fast development is changing. Excited to see where AISheeter goes!
thanks so much for breaking your process down like that. I've been VERY curious about this for some time.. but have not had any time to play around myself.
In terms of the 20% traditional coding, was that something you had to figure out 100% yourself? or can a newbie also be guided on this part as well? I totally understand the benefit of just doing it quickly yourself, especially if you already have something to reuse that works.
This is how I will be teaching my 11yr old son to code going fwd!
re: Ai/Sheets Can it propose and generate a better "P&L" based on feeding it "myShittyAttempt" and the "Ideal_P&L_Sheet"? Would using Vision even help in this instance?
Those 20% can sure be helped by AI , but since I got the coding skills few years back, it was better for me to put in the logic myself,save more time.
Yes, in term of aisheeter, will add more items once I have more requests from users, Vision is a great idea that I could try .
Thanks
This is awesome! Integrating multiple LLMs into Google Sheets sounds super useful, especially for quick AI tasks. How's the performance with larger datasets, and are there any plans to add more AI models down the line?
Thanks, The performance depends on the provider like OpenAI, Claude, Groq and Google (Gemini), so it shouldn't be a problem. Let's see the interests I got the users then I will plan to add more models.