Warpstead’s AI runs on your PC. Your files, your photos and your chats don’t go to any server, and once it’s installed it works offline.
Installing the engine
The engine isn’t installed at first. Open Settings → Engines → AI engine and press Install. It downloads the model and its vision part once, about 1.9 GB, checks them, and keeps them on disk. A notice in the corner follows the download and tells you when the engine is ready.
The model is Qwen3.5 2B, small enough to run on an ordinary PC; it uses the graphics card through Vulkan. Remove in the same place deletes the downloaded files and frees the space.
Without the engine, the AI features in this chapter say so and point you to Settings. Everything else in Warpstead works without it.
What uses it
- Tag with AI in a Folder node: tags and a short description for each file. See Folders and files.
- AI Classify in a pipeline: a category, tags and a subject for each image. See Pipelines.
- The app list (App… in the node picker) sorts your programs into categories. The ones it isn’t sure about are marked to check, and you can right-click any app to set its category yourself. Reclassify apps in Settings starts over.
- AI Chat, below.
AI Chat
Add an AI Chat node and type. Enter sends, Shift Enter starts a new line, Stop interrupts the answer, Clear chat starts over. It answers in the language of the interface, briefly.
Wire a Folder node into the chat and the model gets the names of the files in that folder, so you can ask about them. It sees the names only, not what’s inside the files, and the chat can’t look at images or PDFs.
The engine answers one request at a time: if it’s busy tagging a folder, the chat waits its turn and tells you.
Limits
A 2-billion-parameter model is quick and private, and it makes mistakes. Tags and categories are suggestions: check them before you rename or move files based on them, which is what the preview in pipelines is for.