Configuring Agents

Define and customize agents using JSON: set up personality traits, emotional parameters, memory files, LLM preferences, and voice profiles — all without writing Rust.

📄 Agent Config Overview

Agents are fully defined in a single config.json file.

Example:

{
  "agents": [
    {
      "id": "velma",
      "personality": {
        "curiosity": 0.7,
        "trust": 0.4,
        "anger": 0.1
      },
      "goals": [
        {
          "id": "explore_ruins",
          "priority": 0.8
        }
      ],
      "memories": [
        {
          "content": "Player helped me escape the cave.",
          "tags": ["trust"],
          "importance": 0.9
        }
      ],
      "llm_profile": {
        "preferred_provider": "groq"
      }
    }
  ]
}

---

🧠 Personality

Set the emotional baseline of your agent:

"personality": {
  "curiosity": 0.6,
  "anger": 0.2,
  "happiness": 0.8
}

• Values range 0.0 to 1.0

• These affect tone, memory bias, and behavioral tendencies

---

🎯 Goals

Each goal is:

{
  "id": "explore_cave",
  "priority": 0.9
}

• Goals can be reprioritized during runtime

• The agent’s actions aim to resolve the top goal

---

🗃️ Memory"memories": [

  {
    "content": "Marcus gave me the map.",
    "tags": ["trust", "plot_critical"],
    "importance": 0.7
  }
]

• Importance influences recall likelihood

• Tags link to emotions and behavior trees

---

🔁 Inference Settings

LLM routing is configured per-agent or globally:

"llm_profile": {
  "preferred_provider": "openai",
  "fallbacks": ["groq", "local"]
}

• You can set provider priority or override at runtime

• Local models supported via llm_service.rs

---

💡 Tips

• Use separate JSON files per agent and merge at build time

• Goals and emotions can be modified by gameplay triggers

• For testing, you can simulate config overrides via CLI

---

🔗 Related Pages

• 5. Inference System

• 8. API Reference → config.rs