Private AI Assistant
A private AI assistant for work you can trust: keys in Keychain, capture on demand, inference through the provider you connect.
Reviewed July 21, 2026. GuideLayer publishes this page; we mark our own product clearly and stick to facts we can re-check on each vendor's site.
You can't get better at software without the software in front of you. That means trusting who sees the screen.
Every assistant says the right things in the fine print. Fewer let you check what leaves the machine. That's the real trust gap for screen-aware practice, same tension we cover in our screen-aware AI assistant guide.
GuideLayer is built so you can inspect the posture: keys in Keychain for your own routes, capture on demand, and a visible inference path through the provider you connect. The ability you build stays yours too, an Ability Map stored on the Mac, not a chat history on someone else's server. Contrast with local AI assistant for Mac when you want to stay on your own Codex or Claude API key.
What actually works
Your providers, not ours
Runs on Codex or a Claude API key already on your Mac. GuideLayer doesn't host your API keys on a remote server.
Capture only when you ask
Screen capture happens on demand. Not a continuous feed to anyone's cloud.
Posture you can check
Privacy Center shows memory, screenshots, and routing in one place. One-switch incognito when the work gets sensitive.
Encrypted at rest
Conversations and saved tutorials encrypted on the device. Not plaintext in a cache folder.
Safe enough to practice, not air-gapped theater
GuideLayer isn't a fully offline model you download once and never phone home. By default it can route through your Codex subscription or Claude API key. The bet is still checkable: capture on demand, Ability Map on your Mac, and Privacy Center showing which route is active, not a standing vendor help desk that owns your screen stream. For the product overview, see AI assistant for Mac.
Workflow examples
Sensitive client work on screen
Law firm, finance, health admin. The screen has data you can't paste into a random web chat.
- 01Enable incognito for the session. Screenshots and memory stay off.
- 02Ask by voice. GuideLayer captures context only when triggered.
- 03Review Privacy Center settings before and after the task.
Daily use without overthinking it
You want screen help but don't want to re-read a privacy policy every morning.
- 01Set defaults once: screenshot on demand, encrypted memory on.
- 02Use your existing Codex subscription or Claude API key.
- 03Incognito is there when the work gets sensitive.
Comparing to cloud-only assistants
You're deciding whether screen-aware help is worth the trust tradeoff.
- 01Check where keys live (Keychain vs vendor cloud).
- 02Check when screenshots fire (on demand vs always).
- 03Check if there's a vendor server in the loop beyond the model provider.
Private AI isn't a marketing adjective. It's where your keys live, when your screen gets captured, who can see the conversation besides you, and where the ability you build ends up, start with AI assistant for Mac for the full product tour.
Common questions
Does GuideLayer stream my screen in the background?
No. Screen capture happens on demand when you ask. Inference uses your Codex subscription or Claude API key. Privacy Center shows which path is active.
Is GuideLayer fully offline?
No. It's local-first, not air-gapped. Inference uses your Codex subscription or Claude API key. Privacy Center controls memory, screenshots, and incognito sessions.
Where are my API keys stored?
When you bring your own provider, keys live in the macOS Keychain, device-only. GuideLayer doesn't ask you to paste those credentials into a website form.
Related guides
Try GuideLayer on your Mac
Stuck in an app? Download for Apple Silicon. GuideLayer points to the control you need. Connect the AI you already use.