LLM Helper for Foundry VTT

Chat with locally hosted LLMs inside Foundry VTT, pull context from your Journals, and now generate NPCs directly onto the scene—no external paid APIs required.

Highlights

• Local LLM first: Works with Ollama (default) and other compatible endpoints.
• Modern UI (v13-ready): Uses the ApplicationV2 APIs and the new v13 scene control structure, with legacy fallback for v12.
• Context-aware answers: Lightweight RAG over Journal entries for in-world responses.
• NPC Generator: One-click prompt to create an NPC actor and drop a token on the map.
• Custom personas: Tune the system prompt, model, and context window per world.

Install

1. Foundry VTT → Add-on Modules → Install Module.
2. Manifest URL:

   https://raw.githubusercontent.com/Gizmo3030/LLM-Helper-FoundryVTT/refs/heads/main/module.json
   

3. Install and enable in your world.

Configure

Ollama (recommended)

1. Install Ollama and pull a model, e.g. ollama pull llama3.2.
2. Allow cross-origin requests:
   • Linux/Mac: OLLAMA_ORIGINS="*" ollama serve
   • Windows: set env var OLLAMA_ORIGINS="*" before starting Ollama.
3. Ensure Ollama is running (default http://localhost:11434).

Module settings (Game Settings → Configure Settings → LLM Helper)

• Provider: ollama (or other configured backends).
• API Endpoint: defaults to http://localhost:11434/api/generate (auto-normalized if you give a base URL).
• Model: the pulled model name (e.g. llama3.2).
• System Prompt / Context Window: shape the assistant.
• NPC Template: edit the JSON-format prompt used for NPC generation.
• Click Connect to test.

Use

• Open the LLM control on the left toolbar (brain icon).
• Tools:
  • Open LLM Chat: chat window with journal-aware context.
  • Open LLM Settings: quick access to settings UI.
  • Generate NPC: prompt → actor creation → optional token drop at view center.

Tips / Troubleshooting

• If Ollama rejects requests, re-run with OLLAMA_ORIGINS="*" ollama serve.
• Endpoint flexibility: you can supply base http://host:11434 or full /api/generate; the module normalizes it.
• If JSON parsing fails for NPC generation, check the LLM response formatting and the template you provided.

Compatibility

• Foundry VTT: v12+, verified on v13
• System: system-agnostic (NPC creation assumes D&D 5e fields; adjust template for other systems)

License

MIT

Author

Gizmo3030