Components

Components are the things the app can get for this computer rather than for an AI. The base install stays small; each component downloads only when you ask and says how big it is first.

Get here, manage in Settings

Add-ons › Components is where you get them. Settings › Storage is where you manage them: where models and components live (move the folder to any drive and future downloads follow), what is installed and its size, and Remove.

What is there

• Projects - Your Own AI Build, the helper that lets an AI work in a folder with you and run tools. About 50 MB; also offered the first time you open a project.
• Engines - the bundled engine runs any model on any GPU. Optional engines help specific hardware: the CUDA engine for NVIDIA cards (offered after your first answer, never downloaded without asking), the MLX engine on Apple Silicon (preview). Their controls live in Settings › Engines; the card here shows your machine's status and takes you there.
• Memory, vision and reading - the small on-device models behind memory recall and smart routing, extraction and smart modes, vision for a downloaded chat model, and OCR for scanned documents. The first two are the helper model, which the home page also offers on its own.
• Programs tools need - uv, Git, Node.js, Python, Docker. Green when present; Install shows exactly what would run, then runs it - an official installer, your system's package manager, or your terminal when a password is needed.

Nothing silent

No component installs itself. The offers you see - the CUDA engine after a first answer, Projects when you open a folder, a program a tool needs - are offers, and each shows what it will do before it does it.

The helper offer

From the second launch, the home page has one offer slot under the hero. The NVIDIA engine takes it first when your card qualifies; otherwise it offers the helper - the helper model and the small memory model, installed together with one button, privately, and never as part of the first download. Once installed, or dismissed, the offer is gone. Without the memory half, memory recall, the meaning-based side of routing and the semantic health check are off, and routing keeps questions on your device rather than guessing. See The helper model.