Install and Select Chat Models
Install a local chat model and the model used to search memories.
Follow the bold button names and numbered steps. Technical names are included only when you need to recognize a model or file inside the app.
Open Settings → Models → Chat Models to manage local and connected models.
Understand the two required roles
- Chat model: writes replies and may provide vision, tool use, or reasoning.
- Embedding model: converts memories into vectors so relevant information can be recalled. It does not write the reply.
The active row and badges communicate capabilities:
- Installed: model files are present;
- Active: currently selected for that role;
- Required: needed for a core feature such as semantic memory;
- Vision: can inspect images;
- Tool Use: can use supported tools;
- Reasoning: supports deeper reasoning behavior.
Install a recommended local model
Goal
Install a model appropriate for the computer and make it active for chat.
Before you start
Connect to the internet, verify free disk space, and keep Local Waifu open during the download.
Steps
- Open Settings → Models → Chat Models.
- Compare model size, RAM guidance, and capability badges.
- Select Install.
- Wait for the download and installation to finish.
- Select Set active or Use for chat.
- Send a short test message.
Expected result
The model shows Installed and Active and generates a response.
If something goes wrong
- Check network, proxy/VPN, firewall, and disk space.
- Choose a smaller model if RAM pressure is high.
- A model that fits on disk may still be too large at runtime because context and KV cache also use memory.
- On Windows, run Settings → Hardware → Local AI acceleration → Check after sending a message.
Install the memory model
Install EmbeddingGemma when it is marked Required. Keep the active embedding model installed. Changing it starts re-embedding; memory recall may temporarily rely more heavily on keywords until the process completes.
Remove a model
Use the trash icon on a model that is no longer active or required. The app blocks deletion of the active chat model and protected embedding model. Switch roles first, then delete.