beevr

Inspiration

We showed up to CruzHacks Hack Day ready to build. While I got started on the architecture, my partner walked down to Bookshop Santa Cruz with one mission: find the owner and ask a single question: "If we built this, would you actually use it?"

They said yes.

That was all we needed. We also asked how they track what to reorder. The answer was a spreadsheet, a gut feeling, and a lot of manual emails. That conversation made us realize small businesses are sitting on a goldmine of data they can never actually use. No SQL knowledge. No data team. No time. Just a pile of docs, emails, and spreadsheets that never talk to each other.

We wanted to fix that for every business, not just bookshops.

What it does

Beevr is a shared company brain. You connect your data: documents, emails, Slack, spreadsheets, and your whole team can ask questions across all of it in plain English. Ask "what are my top selling books this week?" or "which inventory is running low?" and get a source-backed answer in seconds.

But Beevr does not stop at answers. Every response has a "Create Agent" button. One click turns any insight into an automated workflow that runs on a schedule, with a visual canvas you can edit and human approval gates before anything gets sent.

How we built it

• Frontend: Next.js 14, Tailwind, shadcn/ui, React Flow
• AI layer: Vercel AI SDK + Claude Sonnet for brain chat, agent generation, and structured tool calls
• Deployment: Vercel, every push auto-deploys
• Data: Real inventory scraped from Bookshop Santa Cruz's live website, seeded as the demo tenant

Challenges we ran into

Getting the agent builder to generate a structured, renderable workflow from a single natural language sentence was the hardest part. We had to carefully prompt Claude to return consistent JSON representing nodes and edges that React Flow could render, and handle all the edge cases where descriptions were too vague or too complex.

We also had 4 hours. So there's that.

Accomplishments that we're proud of

Getting real customer validation the same morning we built it. Bookshop Santa Cruz did not just inspire the demo: they confirmed the problem is real and the solution is something they would actually pay for. Shipping a working brain chat, document indexing, agent builder, and visual workflow canvas in one hackathon session felt pretty good too.

What we learned

The gap between "chat with your docs" and "automate work from your docs" is smaller than we thought. And that gap is exactly where Beevr lives. We also learned that the best validation you can get costs nothing: just walk into a business and ask.

What's next for Beevr

Multi-tenant workspaces so any business can onboard in minutes. A self-hosted version for enterprise teams who cannot put their data in the cloud. And a return visit to Bookshop Santa Cruz to show them what we built.

Built With

• anthropic-claude-api
• next.js
• node.js
• react-flow
• shadcn/ui
• tailwind-css
• typescript
• v0
• vercel
• vercel-ai-gateway
• vercel-ai-sdk