Kept | Devpost

homepage
workspace
what kept catches
promises

Inspiration

Creators make small public promises all the time: “I’ll send the template tomorrow,” “Part two is coming next week,” or “Comment GUIDE and I’ll share the link.” Those promises become scattered across posts, threads, transcripts, and comments—and are easy to forget.

Most social tools help people create more content. Kept focuses on what happens after publishing: helping creators remember what their audience is waiting for and follow through.

What it does

Kept is a private, evidence-backed promise ledger for creators.

Paste a post, thread, transcript, or comment into Kept, and it identifies likely commitments, preserves the exact source quote, highlights what is open or overdue, and prepares an editable follow-up draft.

The creator stays in control. Kept does not connect to social accounts, store social credentials, scrape platforms, or publish automatically. It prepares the next action; the creator decides whether and where to post it.

How I built it

I built Kept with Next.js, TypeScript, Tailwind CSS, Neon Auth, Neon Postgres, Drizzle ORM, and Vercel.

The core scan uses a deterministic promise detector, with optional server-side AI enrichment for ambiguous wording. Each result keeps its original evidence attached, so creators can review the source rather than trust a black box.

Authenticated workspaces use a server-derived identity and Postgres storage, so saved sources, promises, and drafts remain private and can be accessed from another browser or device. The /demo route provides a seeded workspace for judges to experience the full flow without creating an account.

Challenges I ran into

The biggest challenge was avoiding false confidence. Language can sound like a promise without being one, so Kept shows the source quote, makes results editable, and lets creators dismiss anything that is not relevant.

I also had to balance automation with privacy. Instead of asking for social-media passwords or pretending to publish on a creator’s behalf, I kept the MVP text-first and human-controlled.

What I learned

I learned that useful social-media automation is not always about generating more posts. Sometimes the highest-value automation is helping someone keep the commitments they have already made.

I also learned that attaching evidence makes AI-assisted suggestions more trustworthy, easier to verify, and easier to act on.