DevBrief — AI Standup Writer Powered by Hermes Agent (Ver...

Hermes Agent Challenge Submission

This is a submission for the Hermes Agent Challenge

What I Built

DevBrief turns GitHub activity into human-readable standups, PR changelogs, or work logs. Any visitor can sign in with GitHub OAuth, pick a repo, set a time range and branch, filter PRs and authors, choose a tone (casual / formal / concise), and hit Generate.

The Next.js app does not call OpenRouter directly. It fetches commits, PRs, and issues from GitHub, then calls Hermes Agent’s OpenAI-compatible API (POST /v1/chat/completions) on a long-running gateway. Hermes runs the agent loop (skills, tools, server-side model config) and returns the final brief.

Architecture:

Users → DevBrief (Vercel)
            ↓  POST /api/summary → lib/hermes.ts
      Hermes gateway (Docker on Render)
            ↓  model: openrouter/owl-alpha
      OpenRouter (API key only on Hermes — not on Vercel)

Demo

Try it:

Open devbrief-tau.vercel.app
Connect GitHub → select a repo → choose output mode and tone → Generate
Confirm Hermes is up: https://devbrief-hermes.onrender.com/health

Code

Repository: github.com/sundayonah/devbrief

Piece	Location
Hermes client	`lib/hermes.ts` → `HERMES_ENDPOINT` + `POST /v1/chat/completions`
Summary API	`app/api/summary/route.ts`
Standup skill	`hermes-skills/standup-writer.md`
Hermes Docker image	`docker/hermes/Dockerfile`
Production model config	`docker/hermes/config.yaml` (`openrouter/owl-alpha`, `max_tokens: 2048`)
Deploy guide	`docs/DEPLOYMENT.md`

My Tech Stack

Layer	Technology
Frontend	Next.js 14 (App Router), TypeScript, Tailwind CSS
Auth & GitHub	NextAuth.js, Octokit, GitHub OAuth
AI agent	Hermes Agent — `hermes gateway run` + API server
Model	`openrouter/owl-alpha` via OpenRouter (on Hermes host only)
App hosting	Vercel
Agent hosting	Docker on Render

How I Used Hermes Agent

Hermes is not a chatbot wrapper here — the gateway is the brain for every generation.

Hermes capability	How DevBrief uses it
API server	`API_SERVER_ENABLED=true`; Next.js calls `/v1/chat/completions` server-side (no browser CORS)
Gateway	`hermes gateway run` in Docker on Render — not inside Vercel serverless
Skills	`standup-writer.md` copied to `/root/.hermes/skills/` in the image
Server model config	`docker/hermes/config.yaml` sets `model.default: openrouter/owl-alpha` (request `model` field is not what drives inference)
OpenRouter	`OPENROUTER_API_KEY` on Render only — not in Vercel env
Auth	`API_SERVER_KEY` on Render ↔ `HERMES_API_KEY` on Vercel
Cron / messaging	`hermes schedule` documented as a next step in the UI; Slack/Telegram delivery disabled in current deploy

Request flow:

POST /api/summary
  → GitHub API (user OAuth token)
  → generateBrief() in lib/hermes.ts
  → Hermes POST /v1/chat/completions
  → standup / PR changelog / work log
  → UI (copy, edit, history)

The long setup story (WSL PATH, duplicate .env keys, 127.0.0.1 vs localhost, OpenRouter 402, baking config.yaml for Render) is in the tutorial post.

What I learned

Start the gateway, hit /health, then /v1/chat/completions before wiring the app.
Hermes reads ~/.hermes/config.yaml for the real model — env vars and JSON model alone were not enough on Render until we shipped docker/hermes/config.yaml.
Split hosting: serverless Next.js + long-running Hermes elsewhere is the right pattern for this challenge.

Thanks for reading — try the live demo and leave a comment if you hit snags with Hermes on Render or Vercel.

推荐订阅源

DEV Community