Hermes Atlas Ecosystem Map

Hermes Atlas Ecosystem Map

Skill by ara.so — Hermes Skills collection.

What is Hermes Atlas?

Hermes Atlas is a community-curated directory of the Hermes Agent ecosystem—mapping every tool, skill, integration, and deployment template built for Hermes Agent by Nous Research. It features:

• Quality-filtered repository catalog (80+ projects across 12 categories)
• Live GitHub star counts with 30-day sparklines
• RAG-powered chatbot ("Ask the Atlas") grounded in 27 research files
• Hybrid search/filter with trending badges
• Zero-framework frontend (vanilla HTML/CSS/JS)
• Serverless API (Vercel functions + Redis cache)

Installation

git clone https://github.com/ksimback/hermes-ecosystem.git
cd hermes-ecosystem
npm install

Dependencies (only 2):

• openai — for embeddings and OpenRouter API
• redis — for star count caching and history

Key Files

hermes-ecosystem/
├── index.html                 # Main map UI (single-page app)
├── data/
│   ├── repos.json            # Single source of truth (84 repos)
│   └── chunks.json           # Pre-computed embeddings (283 chunks, 7MB)
├── api/
│   ├── stars.js              # Live star counts (1hr cache)
│   ├── stars-history.js      # 30-day sparkline data
│   └── chat.js               # RAG chatbot with streaming
├── scripts/
│   ├── build-chunks.js       # Rebuild embeddings from research/
│   └── test-rag.js           # RAG quality tests (27 test cases)
├── research/                  # 27 knowledge base files
└── lib/redis.js              # Shared Redis client

Adding a New Project

1. Quality Filter Criteria

Before adding to data/repos.json, verify:

• Built for Hermes Agent (not generic AI tools)
• Created after July 22, 2025 (Hermes Agent launch)
• Shows genuine effort (not personal pet projects)
• Passes security review (check for credential leaks, malicious code)

2. Add to repos.json

{
  "categories": {
    "skills": [
      {
        "name": "hermes-skill-example",
        "url": "https://github.com/username/hermes-skill-example",
        "description": "Brief description of what the skill does",
        "tags": ["tag1", "tag2"]
      }
    ]
  }
}

Available categories:

• skills — Core agent skills
• tools — External tool integrations
• plugins — Hermes plugins
• deployments — Hosting/infrastructure templates
• frameworks — Multi-skill orchestration
• integrations — Third-party service connectors
• forks — Modified Hermes Agent versions
• ui — User interfaces and dashboards
• data — Datasets and evaluation tools
• research — Academic papers and experiments
• templates — Boilerplates and starters
• misc — Everything else

3. Verify the Change

# Check JSON syntax
node -e "require('./data/repos.json')"

# Preview locally
open index.html

Rebuilding the Knowledge Base

After updating files in research/, rebuild embeddings:

# Set API key
export OPENROUTER_API_KEY=sk-or-v1-...

# Rebuild chunks.json (splits + embeds)
node scripts/build-chunks.js

# Test RAG quality
node scripts/test-rag.js

build-chunks.js does:

1. Reads all .md files in research/
2. Splits into 500-char chunks with 100-char overlap
3. Embeds each chunk using OpenAI text-embedding-3-small
4. Writes to data/chunks.json (static file, committed to repo)

Output:

Processing research/agent-architecture.md...
Processing research/hermes-skills.md...
...
✅ Built 283 chunks (7.2 MB)

Testing the RAG Pipeline

export OPENROUTER_API_KEY=sk-or-v1-...
node scripts/test-rag.js

Sample test:

// scripts/test-rag.js excerpt
const tests = [
  {
    query: "How do I create a custom Hermes skill?",
    expectedKeywords: ["skill.md", "yaml", "frontmatter", "triggers"]
  },
  {
    query: "What's the difference between a skill and a tool?",
    expectedKeywords: ["skill", "tool", "integration", "plugin"]
  }
];

Output:

Test 1/27: How do I create a custom Hermes skill?
✅ Found 4/4 keywords in context
...
27/27 passed (100%)

API Endpoints

GET /api/stars

Returns live star counts for all repos in repos.json.

Caching:

• 1hr TTL in Redis
• Falls back to direct GitHub API if cache miss
• Rate limit: 5000/hr with GITHUB_TOKEN, 60/hr without

Response:

{
  "username/repo": 142,
  "anotheruser/hermes-skill": 89
}

Usage in frontend:

const response = await fetch('/api/stars');
const stars = await response.json();
document.querySelector('[data-repo="username/repo"]').textContent = stars['username/repo'];

GET /api/stars-history?repos=user/repo1,user/repo2

Returns 30-day star count history for sparklines.

Response:

{
  "user/repo1": [120, 122, 125, 128, 130, ...],
  "user/repo2": [45, 47, 49, 51, 53, ...]
}

POST /api/chat

RAG chatbot endpoint with streaming support.

Request:

{
  "message": "How do I deploy Hermes Agent?",
  "conversationHistory": [
    {"role": "user", "content": "What is Hermes Agent?"},
    {"role": "assistant", "content": "Hermes Agent is..."}
  ]
}

Response (streaming):

data: {"type":"context","chunks":[{"text":"...","file":"deployment.md"}]}
data: {"type":"token","content":"To deploy"}
data: {"type":"token","content":" Hermes Agent"}
data: {"type":"done"}

Retrieval pipeline:

1. Conversation-aware rewrite — expands query using chat history
2. Hybrid search — BM25 (keyword) + cosine similarity (semantic)
3. MMR re-ranking — maximal marginal relevance to reduce redundancy
4. Top-K selection — retrieves 5 most relevant chunks
5. LLM generation — Gemma 4 31B with fallback chain

Configuration

Environment Variables (Vercel)

# Required
OPENROUTER_API_KEY=sk-or-v1-...
REDIS_URL=redis://default:password@host:port

# Optional
GITHUB_TOKEN=ghp_...                              # Fine-grained PAT (public read-only)
OPENROUTER_MODEL=google/gemma-4-31b-it:free       # Primary LLM
OPENROUTER_FALLBACK_MODELS=google/gemma-4-26b-it:free,google/gemini-3-flash-1.5

vercel.json Configuration

{
  "functions": {
    "api/**/*.js": {
      "maxDuration": 30
    }
  },
  "crons": [
    {
      "path": "/api/stars-daily-snapshot",
      "schedule": "0 0 * * *"
    }
  ]
}

Common Patterns

Adding Multiple Projects at Once

// scripts/batch-add.js (create this if needed)
const fs = require('fs');
const repos = require('../data/repos.json');

const newProjects = [
  { name: "hermes-skill-web-search", url: "https://github.com/...", category: "skills" },
  { name: "hermes-tool-calendar", url: "https://github.com/...", category: "tools" }
];

newProjects.forEach(proj => {
  repos.categories[proj.category].push({
    name: proj.name,
    url: proj.url,
    description: "",  // Fill in manually
    tags: []
  });
});

fs.writeFileSync('./data/repos.json', JSON.stringify(repos, null, 2));

Security Review Checklist

// scripts/security-check.js (conceptual)
const checks = [
  {
    name: "No hardcoded credentials",
    test: (code) => !/api[_-]?key\s*=\s*["'][^"']+["']/i.test(code)
  },
  {
    name: "No eval() usage",
    test: (code) => !/eval\(/.test(code)
  },
  {
    name: "Dependencies up to date",
    test: async (repo) => {
      // Check package.json for known vulnerable versions
    }
  }
];

Custom Sparkline Rendering

// From index.html (adapted for skill documentation)
function renderSparkline(history) {
  const max = Math.max(...history);
  const min = Math.min(...history);
  const range = max - min || 1;
  
  const points = history.map((val, i) => {
    const x = (i / (history.length - 1)) * 100;
    const y = 100 - ((val - min) / range) * 100;
    return `${x},${y}`;
  }).join(' ');
  
  return `<svg viewBox="0 0 100 100"><polyline points="${points}" /></svg>`;
}

Troubleshooting

Embeddings Build Fails

Error: Error: 429 Too Many Requests

Fix: Add rate limiting to build-chunks.js:

async function sleep(ms) {
  return new Promise(resolve => setTimeout(resolve, ms));
}

for (const chunk of chunks) {
  const embedding = await getEmbedding(chunk.text);
  chunk.embedding = embedding;
  await sleep(100); // 10 req/sec max
}

Redis Connection Timeout

Error: ECONNREFUSED or ETIMEDOUT

Fix: Check REDIS_URL format:

# Correct format
redis://default:password@host.region.cloud.redislabs.com:12345

# Common mistake (missing default user)
redis://:password@host...

Stars Not Updating

Error: Cached star counts stale after 1 hour

Fix: Manually bust cache:

# Via Redis CLI
redis-cli -u $REDIS_URL
> DEL stars:cache

Or programmatically:

// api/stars.js — force refresh
if (req.query.refresh === 'true') {
  await redis.del('stars:cache');
}

RAG Returns Irrelevant Results

Symptom: Chatbot answers unrelated to Hermes Agent

Fix: Check chunk quality:

node scripts/test-rag.js --verbose

# Review chunks.json for short/low-quality chunks
node -e "
const chunks = require('./data/chunks.json');
const short = chunks.filter(c => c.text.length < 200);
console.log('Short chunks:', short.length);
"

Increase chunk size in build-chunks.js:

const CHUNK_SIZE = 800;  // was 500
const OVERLAP = 150;     // was 100

GitHub Rate Limit Exceeded

Error: 403 rate limit exceeded

Fix: Add GITHUB_TOKEN to Vercel env vars:

1. Go to GitHub → Settings → Developer settings → Personal access tokens → Fine-grained tokens
2. Create token with Public repositories (read-only) permission
3. Add to Vercel: GITHUB_TOKEN=ghp_...

This increases rate limit from 60/hr to 5000/hr.

Local Development

# Open the map in a browser (API endpoints won't work locally)
open index.html

# Test API endpoints locally (requires env vars)
export OPENROUTER_API_KEY=...
export REDIS_URL=...
vercel dev

Note: The frontend is pure static HTML—no build step required. API endpoints only run on Vercel or with vercel dev.

Deployment

# Deploy to Vercel
vercel --prod

# Set environment variables
vercel env add OPENROUTER_API_KEY
vercel env add REDIS_URL
vercel env add GITHUB_TOKEN

The daily cron job (/api/stars-daily-snapshot) automatically runs at midnight UTC to populate sparkline history.

Live site: hermesatlas.com
Repository: github.com/ksimback/hermes-ecosystem