@artale/pi-hermes

Self-improving agent for Pi.

Inspired by Hermes Agent (107K stars) + Hermes Agent Self-Evolution (2K stars).

Features

Memory

// Remember important things
hermes.remember({ content: "User prefers dark mode" })

// Recall later
hermes.recall({ query: "preferences" })

Skill Creation

// Create skill from experience
hermes.learn({ 
  task: "fix authentication bug", 
  outcome: "Added token validation",
  quality: 8  // 1-10
})

Self-Evolution (GEPA)

// Evolve skill based on feedback
hermes.evolve({
  skill: "fix-bugs",
  feedback: "Add more test cases"
})

User Modeling (Honcho-style)

// Track preferences
hermes.model({
  "format": "json",
  "verbose": "brief"
})

// Get user model
hermes.model("get")
// Returns: { preferences, feedback, traits }

Installation

pi install npm:@artale/pi-hermes

Commands

| Command | Description | |---------|-------------| | remember | Store important memories | | recall | Search memories | | learn | Create skill from task outcome | | evolve | GEPA-style optimization | | model | Build user model | | skills create | Manual skill | | use | Use skill |

Comparison

| Feature | Hermes Agent | pi-hermes | |---------|-------------|----------| | Memory | ✅ | ✅ | | Skill creation | ✅ | ✅ | | Self-evolution | ✅ (external) | ✅ | | GEPA | ✅ | ✅ | | User modeling | ✅ (Honcho) | ✅ | | Multi-platform | Telegram, Discord | CLI only |

MVI Approach

• Single file implementation
• No external dependencies
• JSON file storage
• CLI-first

License

MIT