npx minionsai

minions --version
npm view minionsai version

undefined

undefined

// Tasks are displayed in real-time Kanban columns
// Watch live streaming output in the task detail view
// Completion judge runs after each agent turn

// Example: Daily website monitoring routine
{
  name: "Check competitor pricing",
  schedule: "0 9 * * *", // 9 AM daily
  prompt: "Visit competitor.com and extract current pricing, compare to yesterday",
  workspace: "./data/competitive-intel"
}

// Task prompt:
"Research the top 10 AI coding assistants launched in 2025. 
Create a comparison table with: name, pricing, key features, GitHub stars.
Save as markdown in research/ai-coding-tools.md"

// Agent will:
// 1. Search for AI coding assistants
// 2. Visit documentation sites
// 3. Gather pricing and feature data
// 4. Format as markdown table
// 5. Save file and propose completion

// Routine configuration:
{
  name: "Daily HackerNews trending",
  schedule: "0 */6 * * *", // Every 6 hours
  prompt: `
    1. Fetch top 30 HackerNews stories
    2. Extract: title, points, comments, URL
    3. Append to data/hn-trending.jsonl with timestamp
    4. If any story mentions 'AI agents', save separately to data/hn-ai-agents.jsonl
  `,
  workspace: "./scraped-data"
}

// Task prompt:
"Create a TypeScript function that:
- Takes an array of URLs
- Fetches each in parallel with retry logic
- Returns success/failure status for each
- Includes proper error handling and types
Save to src/utils/fetch-urls.ts with tests"

// Monitor live as agent:
// - Creates the implementation
// - Adds TypeScript types
// - Writes tests
// - Runs them to verify

// Multi-step task:
"1. Read all markdown files in content/drafts/
2. For each draft:
   - Check grammar with language tool
   - Generate 3 title alternatives
   - Create social media preview text
   - Move to content/ready/ when done
3. Create summary report in content/processing-report.md"

undefined

undefined

// Minions creates Hermes sessions via API
// Each task = one persistent Hermes root session
// Sessions persist across Minions restarts
// Chat history stored in Hermes's database

// Task metadata (Minions SQLite):
{
  id: "task_123",
  title: "Research AI tools",
  status: "in_progress",
  hermesSessionId: "sess_abc", // Reference to Hermes
  model: "claude-3-5-sonnet",
  reasoningEffort: "medium"
}

// Chat transcript: stored in Hermes session database
// File outputs: in workspace directory

undefined

Example workspace structure:

undefined

undefined

undefined

undefined

undefined

// Use the UI to:
// 1. View live streaming output
// 2. Check if agent is waiting for input
// 3. Manually intervene via chat
// 4. Or cancel and restart the task

// Ensure LLM API key is set:
// export OPENAI_API_KEY=your_key

// Check Settings page for completion judge configuration
// Verify API quota/rate limits

undefined

undefined

// ✅ Good: Specific, measurable outcomes
"Scrape product prices from 5 e-commerce sites, 
save to products.json with timestamp, 
compare to last week's data in products-history.json"

// ❌ Bad: Vague, no clear completion criteria
"Research some products"

undefined

undefined

// Use cron syntax for schedules
"0 9 * * 1-5"   // 9 AM weekdays
"*/30 * * * *"  // Every 30 minutes
"0 0 * * 0"     // Weekly on Sunday

// Build in error recovery:
"If website is down, retry 3 times with 5-minute delays.
Log failures to errors.log"

// In task prompt, specify completion:
"Task is ONLY complete when:
1. All 50 URLs have been processed
2. results.json exists and is valid JSON
3. error_count in summary.md is 0
4. All output files are committed to git"

// Create dependent tasks:
// Task 1: "Scrape data, save to data/raw.json"
// Task 2: "Read data/raw.json, analyze, create report.md"
// Task 3: "Read report.md, generate executive summary"

// Monitor all on Kanban board
// Each progresses independently