APHOTICDocumentation / Why
Why Aphotic
One machine, doing four different jobs: gaming, hosting local AI models, day-to-day development work, and security research. That’s the problem Aphotic exists to solve, and it’s worth explaining why that combination doesn’t just work by default.
The collision
Windows fences off memory and driver claims between applications more or less by default. Linux doesn’t hand you that for free. Start a large local model and launch a game on the same box, and both grab VRAM with nothing arbitrating between them. Individually, either workload runs fine. Run them together and one of them dies, or the whole session does, because nothing on the desktop noticed the collision coming.
That’s not a hypothetical. It’s the exact failure this project was built around: a long session with a local model running, or Claude open in a terminal, and a game launched on a second monitor while the rest of the system keeps working. Two demanding workloads on one GPU, and no desktop shell watching for the fight before it starts.
A desktop that notices
The Resource Engine is the answer: when a game and a local model both want the GPU, Aphotic detects the contention and asks which one yields, instead of letting one crash. It never decides for you and it never kills a process outright. The goal is a game running on one monitor while the rest of the desktop, including a model or an agent session, stays fully functional and isolated from that contention rather than fighting over it. See that page for exactly what’s shipped today versus what’s still ahead.
Security research, without a second machine
The same reasoning applies to the fourth identity. Running a full separate Kali install just to do security research or CTF work is a lot of overhead for something that should be an opt-in mode on the same desktop, not a second operating system. Aphotic’s exploit layer ships as focused, opt-in sublayers (recon, web, network, passwords, wordlists, reversing, forensics, reporting) behind an explicit authorized-use disclaimer, rather than one bundled everything-on install. See Security for the full breakdown.
That sublayer split has a second purpose beyond keeping installs lean: it’s the shape a fully detached, isolated research environment needs eventually. Splitting security tooling into named, independently-selectable pieces now is what makes isolating that side of the desktop later a scoping problem instead of a rewrite.
Modular by design, not by accessory
None of this is worth much if it forces AI tooling on someone who wants none of it. Aphotic’s plugin and layer system means a feature is either installed and gated on, or it doesn’t exist on your desktop at all. Skip the agent plugins and the agent bar icon, Agent Graph, and every AI-tracking surface are absent, not hidden behind a setting you have to go find and turn off. Skip the ai layer and nothing local (Ollama, the Assistant) is installed or offered. The Command Center’s AI Chat tab is the one AI surface in the core shell; without the ai layer it only lists the providers that need no local install (Claude through its CLI, Gemini, ChatGPT), and each stays unavailable until you configure it. The same rule holds for gaming, development tooling, and security research: every identity is opt-in, and staying out of one costs you nothing.
See Profiles & Layers for how the install-time choices work, and Plugin System for how a specific feature gets added or removed after the fact.