What it does Hermes Atlas is a community-curated directory of tools, skills, and integrations for Hermes Agent, Nous Research’s self-improving AI agent. It filters 84 repositories across 12 categories, tracks live GitHub stars with sparklines, and runs a RAG chatbot that answers questions grounded in 27 research files. The whole thing lives at hermesatlas.com.

The interesting bit The frontend is vanilla HTML/CSS/JS — no build step, no framework, no Vite, no tears. The “backend” is three Vercel serverless functions talking to Redis Cloud and OpenRouter. The RAG pipeline is over-engineered in the best way: hybrid BM25 + cosine similarity, MMR re-ranking, conversation-aware query rewriting, and a fallback chain of three Gemma models. All embeddings are pre-computed at build time and served as static JSON. It’s a 2025 architecture wearing 2006 clothes.

Key highlights

• 84 repos, security-reviewed, with live star counts cached in Redis (1hr TTL)
• RAG chatbot with 27 research files, streaming responses, and 27/27 local test pass rate
• Zero frontend dependencies — open index.html and you’re running
• Daily cron job snapshots star history for sparklines and trending badges
• Just two npm dependencies: openai and redis

Caveats

• API endpoints only work when deployed to Vercel; local preview is static-only
• The “February 2026” launch date for Hermes Agent appears to be a typo or forward-dated claim in the README
• Requires four separate API keys/services to run fully (GitHub, OpenRouter, Redis, OpenAI for embeddings)

Verdict Worth studying if you’re building lightweight directory sites or want to see how far vanilla JS + serverless can stretch. Skip it if you need a framework’s ecosystem or are allergic to managing API keys.