Columny

• 

  LOGO

Inspiration

I used to run email outreach campaigns for B2B agencies, and honestly, it was a nightmare. Tracking unique metrics for different prospects across various countries in static Excel sheets was incredibly painful and error-prone.

I kept thinking: what if I could just dump my raw, messy notes into an AI, and it magically structured everything into a clear, visual format?

We realized people everywhere were wasting countless hours doing manual data entry in bloated software. That exact frustration is why we built Columny.

What it does

Columny is a flexible agentic workspace. It isn't a rigid CRM; it adapts entirely to your needs.

Dynamic Data Structuring You can type absolute gibberish into the chat, and the AI instantly turns it into beautiful, structured data.

Infinite Use Cases Whether you are tracking B2B sales performance, logging staff updates, or you are a YouTube creator comparing thumbnail formats against editing styles, Columny handles it.

Total Autonomy The AI autonomously extracts entities, builds your database schema on the fly, and generates real-time dashboard charts so you can visually track anything without ever touching a spreadsheet cell.

How we built it

The Core Stack We used React and Vite for a lightning-fast UI, relying on Supabase for our real-time Postgres syncing and database architecture.

The Agentic Engine The heavy lifting happens in our Supabase Edge Functions paired with the Gemini API. This integration handles our intent routing and executes dynamic SQL commands on the fly.

Desktop AI Integration We also built an official Model Context Protocol (MCP) server so users can securely query their live Columny data from external desktop AI agents.

Challenges we ran into

The Cost Bottleneck Our absolute biggest roadblock was cost. Early on, we were burning through cash using massive, expensive AI models to handle the complex database reasoning. We were worried the project wouldn't be financially viable to scale.

The Pivot to Gemini 2.5 Flash We had to completely rethink our prompting architecture and strict JSON routing to adapt our workflow to a smaller, faster model like Gemini 2.5 Flash. Getting a lightweight model to flawlessly execute dynamic SQL without hallucinations took late nights, but getting it to work effortlessly felt amazing.

Accomplishments that we're proud of

True Agentic Behavior We moved beyond treating AI as a simple text generator and successfully turned it into an autonomous Database Administrator. Watching the AI autonomously create a new column on the dashboard just because you mentioned a new type of data point feels like absolute magic.

What we learned

We learned that the true magic of LLMs isn't in generating chat responses; it is in structured outputs. By forcing the AI into strict JSON schemas and routing user intents, you can turn a standard text model into a highly reliable, autonomous database administrator.

What's next for Columny

Omni-Format Scraping: Currently, we only support text input. Our immediate next step is allowing users to upload PDFs, CSVs, JPGs, and DOCX files so the AI can autonomously scrape and structure data from any format. Agentic Enrichment: We plan to introduce multi-agent workflows to automatically cross-reference and enrich the data you log. Automated Webhooks: Allowing the Columny agent to autonomously trigger Zapier workflows or send follow-up emails based on the data it extracts. Voice Integration: Allowing field sales reps to simply dictate their day into the Columny app and having the agent structure the entire day's CRM data instantly.

Built With

• css3
• gemini-api
• html5
• javascript
• model-context-protocol
• node.js
• postgresql
• react
• supabase
• vite