Frequently Asked Questions

Why does aiMultiFool need Ollama?

Previously, aiMultiFool included its own built-in LLM inference engine using C# and Llamasharp. While powerful, it was difficult to maintain. We’ve simplified things—aiMultiFool is now a lightweight, Python-based front end for Ollama, making everything faster, cleaner, and easier to update.

How does the new Vector Chat tool work? It seems shhh magic!

Vector Chat uses Qdrant, a vector database, to store and retrieve past chats as embeddings—mathematical representations of text. These embeddings are created using a small worker model running inside Ollama.

This enables a technique called retrieval-augmented generation (RAG), where similar past messages are pulled in to give context to new ones. This makes the AI feel more consistent and personal over time—it “remembers” you.

Vector Chat Workflow

• You send a message.
• The app creates an embedding.
• Qdrant searches for similar past exchanges.
• Those are sent to Ollama along with your message.
• Ollama responds using the retrieved context
• The new exchange is stored as a vector in Qdrant.

How can Vector Chat import a novel? Have you lost your mind?

This is an experimental feature still in early development. When you import an EPUB file or book, we scan for chapters and paragraphs, breaking the book into manageable chunks and turning them into vectorized entries. While it’s not perfect yet, it gives the AI searchable context to simulate the theme and tone of the book.

Why does the AI get confused about which role it’s playing in my scenario?

When using the Card Tools’ Action Menu to create a new scenario, make sure to set the first_mes field—the first line your character says. This is crucial for establishing your role.

Example:
"Barman, give me a whiskey please."
Without it, the AI may try to control the scene instead of letting you take the lead.

Why aren’t the export or play buttons working after I create a custom scenario JSON?

Sometimes, the AI might produce an invalid JSON due to formatting errors or lazy outputs. If this happens, go to Card Tools → Fix Character Card JSON to automatically repair it.

This issue is more likely to occur with tired or overloaded models. Starting fresh with a clean model and a new chat usually prevents it.

What’s the fastest way to create and edit Character Cards with AI help?

Use a small, fast model—like a 3B parameter model—for building and editing your Character Cards. Once everything is set up, save the card and switch to a larger, more powerful model for actual roleplaying. It saves time and keeps things smooth.