What I Built

I'm a 17-year-old first-year Computer Science student, and I built StartupForge solo for this hackathon — my first real agentic AI project.

StartupForge is an AI launch-kit autopilot. You describe your product idea in one sentence, and a Hermes Agent runs a four-step launch pipeline, streaming every line live so you review and approve each step before the next one starts:

Competitor Research — live web search: real competitors, pricing, weaknesses, and the gap your product fills.

Landing Page — a complete, responsive HTML page with a working waitlist form, live device preview, and ZIP download.

Launch Posts — an X/Twitter thread, a "Show HN" post, and a Reddit post, each tuned to that platform's culture.

AI Agent Prompts — a build plan plus copy-paste-ready prompts for coding, design, and marketing agents.

The problem: going from "I have an idea" to "I have something to launch" normally takes days of scattered research, copywriting, and planning. StartupForge compresses that into one guided run — with a human-in-the-loop approval gate so you stay in control instead of getting a black-box dump.

My friend 0xwagmi helped me with the CSS and the landing page when I was running low on time — shoutout to him for jumping in.

Code

Geek-Cheth / hermes-agent

Next.js dashboard for Hermes agent product research and launch assets

StartupForge

Describe your idea. The agent does the rest.

Hermes-powered launch kit: describe a product idea in plain text, and the agent autonomously researches competitors, writes a landing page, drafts launch posts, and generates AI agent prompts.

Stack

Next.js 14 (App Router, TypeScript, Tailwind)
Clerk — authentication and per-user run history
Hermes Agent — OpenAI-compatible API at /v1/chat/completions (port 8642)
Supabase — run storage + waitlist
Coolify — reverse proxy on DigitalOcean VPS

Quick start (local)

Copy env file:

cp .env.example .env

Run Supabase schema in your project SQL editor:

# See supabase/schema.sql

Install and run:

npm install
npm run dev

Open http://localhost:3000.

Environment variables

Variable	Description
`NEXT_PUBLIC_CLERK_PUBLISHABLE_KEY`	Clerk publishable key
`CLERK_SECRET_KEY`	Clerk secret key
`HERMES_URL`	Hermes gateway URL (e.g. `http://hermes-agent:8642`)
`HERMES_API_KEY`	`API_SERVER_KEY` from Hermes
`NEXT_PUBLIC_APP_URL`	Public app URL for landing page form POSTs
`SUPABASE_URL`	Supabase project URL
`SUPABASE_SERVICE_KEY`	Supabase service role key
`TELEGRAM_BOT_TOKEN`	Optional fallback Telegram bot

…

View on GitHub

My Tech Stack

Next.js 14 + TypeScript, Tailwind CSS, Framer Motion, Clerk (auth), Supabase (Postgres), SSE streaming, Telegram Bot API, deployed via Docker + Coolify on DigitalOcean. AI: Hermes Agent, powered by MiniMax (I got their Pro plan free through a hackathon, which made running the agent possible).

How I Used Hermes Agent

Hermes is the engine behind every output. I leaned on its agentic capabilities:

Live web search for real competitor data — not hallucinated.

Streaming with live progress so you watch the agent think instead of a spinner.

Multi-step pipeline where the research is fed forward as context into the next three tasks for consistent positioning.

Structured output via a sentinel protocol, parsed into saved, typed results.

Human-in-the-loop approval between every step.

This was an agent problem, not a completion problem — it had to research the real world, produce several different artifacts, carry context between them, and report progress — all of which Hermes handles natively.

Future Improvements

Automated image generation — auto-generate logos, OG images, and landing-page visuals per run.

Dedicated backend — move the in-process job queue to a shared store (Redis/Supabase) for horizontal scaling.

I haven't built these yet simply because I don't have the capital for the extra infrastructure and image APIs right now. As a student (broken guy lmao) , I focused on shipping a solid core experience first.

DEV Community

Hermes Agent Challenge