Concept app · v1.0.0 · Notarized

entAI

Your company AI on the Mac — local first. entAI answers on the Mac's own model whenever it can, and only sends requests to your organization's OpenAI-compatible service (Azure OpenAI, OpenAI, or a gateway) when the local model can't handle it, as your policy allows.

Concept app. entAI explores whether a Mac can run most AI requests locally and only ask a company AI when it truly can't. Routing, packaging, notarization and the local model are verified end to end, including live cloud escalation against a real OpenAI-compatible server — but Microsoft Entra ID sign-in hasn't been tested against a real tenant yet, and the Azure OpenAI backend hasn't been exercised against a live Azure resource either. It's a first release, tested in a demo environment only, not in production, and not a supported product.

Download entAI (2.4 GB) Read the Docs

entAI chat window answering a question on-device, with an On this Mac · Gemma 4 E2B (text) source badge

Features

Local first, cloud when it earns it

💬

Local-First Chat

A menu bar chat window with history and file/image attachments. Every answer shows exactly where it came from — this Mac, or the company AI.

⚡

Quick Chat

Double-press ⌥ anywhere for a floating, Spotlight-style chat panel — no need to switch apps or find the menu bar icon.

🖱️

Right-Click Actions

Select text in any app → Services → entAI: Summarize, Translate, Rephrase, Fix Grammar, Explain, and more — with a live preview and one-click Replace.

📁

File Actions

Right-click a file in Finder to summarize it or ask about it — it opens straight into the chat so you can follow up.

🔀

Deterministic Routing

Too long, contains images, or no local model fits in free memory right now → escalates to the company AI, exactly as your CloudPolicy allows. The model never decides where it runs.

🔒

Consent Before Cloud

Never, Ask, or Automatic — an explicit "Ask Company AI" always counts as consent, and every answer carries a source badge, never silently.

🔑

Any OpenAI-Compatible Backend

API key, Azure OpenAI, or Microsoft Entra ID (OAuth + PKCE, silent via the Enterprise SSO plug-in). No secrets ever stored in configuration profiles.

🛠️

Admin-Defined Actions

Add your own right-click actions — custom prompt, icon, routing, input type — alongside the 12 built-ins, deployed entirely by MDM.

📊

Usage Reporting

Local vs. cloud request counts — never content — via a Jamf Extension Attribute or Intune custom attribute, and right in Settings → Usage.

Status

Where things stand

System Requirements

macOS 26 or later, Apple Silicon only.

8 GB RAM minimum — the on-device model needs about 2.7 GB free.

Build Status

Version 1.0.0. Signed with a Developer ID Application certificate and notarized by Apple.

~2.6 GB installed, including the bundled Gemma 4 E2B on-device model.

What's Verified

Local routing, actions, packaging, signing and notarization, and Services integration are all verified — including live cloud escalation against a real OpenAI-compatible server.

Known Limits

Microsoft Entra ID sign-in has not been tested against a real tenant yet (unit-tested and mock-server only). The Azure OpenAI backend itself hasn't been exercised against a live Azure resource either — real-world cloud testing so far used a different OpenAI-compatible server.