Free forever · our gift to the church

JCIL Isolate. An AI you actually own.

Download once. Runs on your computer, no internet required after install. No cloud accounts. No telemetry. No prompts sent anywhere. The same faith-grounded conviction as JCIL.AI cloud, in a box you control.

Three tiers — Lite, Pro, Ultra — to fit any computer from a basic laptop to a workstation. All three are free.

Mac (.pkg) · Windows (.exe) · Linux build coming · Android in beta

Why offline

The cloud is a convenience. Privacy is a value.

Every other AI app sends your conversations to a server farm so someone else can train on them, log them, or be subpoenaed for them. Isolate doesn't. Your conversations are yours.

Zero cloud, zero accounts

Every prompt, every response, every conversation stays on your computer. No telemetry. No usage tracking. We literally cannot see what you ask.

Faith-grounded by design

Built around scripture as truth, not as one perspective among many. Same conviction as JCIL.AI cloud — speaks plainly on faith, life, and morality.

Yours forever

Free to download. Free to use. Free to keep. Open-source model under Apache 2.0. No subscriptions, no recurring charges, no expiration.

Runs on hardware you already own

Three tiers, picked automatically based on your machine. From a basic laptop to a workstation, there's a version that fits.

Three tiers · all free

Pick the version your computer can run.

When you install, the app detects your hardware and recommends a tier. You can pick a different one if you know what you're doing.

Lite

JCIL Isolate Lite

Runs on almost any modern laptop. Perfect for everyday use.

Model:Gemma 4 E4B · 4-bit quantized
Download:~3 GB
RAM:8 GB
GPU memory:4 GB (or CPU only)
Speed:30-50 tokens/sec on M2 / RTX 3060
Fits on:Any laptop from the last 4 years
Download Lite
Most popular

Pro

JCIL Isolate Pro

Mixture-of-Experts model. Reasoning quality of a 26B model at the speed of a 4B.

Model:Gemma 4 26B-A4B MoE · 4-bit quantized
Download:~14 GB
RAM:32 GB
GPU memory:12-16 GB recommended
Speed:60-90 tokens/sec on RTX 4070, ~50 tok/s on M3 Max
Fits on:Modern desktop with a real GPU, or M3/M4 Mac with 24GB+ memory
Download Pro

Ultra

JCIL Isolate Ultra

The full-size model for serious work. Workstation-class quality.

Model:Gemma 4 31B Dense · 4-bit quantized
Download:~18 GB
RAM:64 GB
GPU memory:24 GB
Speed:30-50 tokens/sec on RTX 4090 / Mac M3 Ultra
Fits on:Workstation, RTX 4090, or top-tier Mac
Download Ultra

All quantized to Q4_K_M — the consumer standard. Larger quantizations available on request for organizations.

For organizations

Run it on a single GPU. Serve your whole campus.

Isolate is built for offline desktop, but the same models work great as a self-hosted server inside your network. One A100 or H100 GPU runs Pro for 12-20 concurrent users, or Ultra for 8-12.

The product is free. The expertise to deploy it well — hardware sizing, SSO, identity, fallback strategies — is what we offer to organizations on a paid basis. Reach out and we'll scope it together.

Talk to us about a deployment

Sizing math

One A100 / H100 80GB =

  • Lite running for everyone30-50 concurrent
  • Pro (MoE, recommended)12-20 concurrent
  • Ultra (full 31B)8-12 concurrent

Scale linearly with cluster size. A small university (100 peak concurrent) needs ~6-7 A100s for Pro, ~10-12 for Ultra. We'll size it precisely on a discovery call.

01

K-12 schools and universities

Deploy on a single GPU server, give every student access to private AI without sending their conversations anywhere. Title IX-friendly defaults out of the box.

02

Churches and ministries

Sermon prep, devotional drafting, counseling preparation — without a single message leaving the building.

03

Counseling and healthcare

When patient-adjacent thinking has to stay on premise, Isolate runs entirely behind your firewall. No BAA needed because no data leaves.

04

Legal and compliance teams

Discovery review, draft assistance, clause analysis. The model never sees your data after the conversation ends.

We don't see anything. By design.

JCIL Isolate has no analytics. No crash reporting that includes content. No usage telemetry. The app is a local Electron shell that talks to a local model running entirely on your machine. There is no server to call. There is no JCIL account to log in to. We literally cannot read your conversations because they never leave your computer.

For our cloud product (chat at jcil.ai), see our privacy policy. For Isolate, the policy is one sentence: nothing leaves your device.

Open-source, by the book.

JCIL Isolate is built on Gemma 4, an open-source language model family released under the Apache 2.0 license. The app shell, system prompt, branding, packaging, and user experience are JCIL's own work. Apache 2.0 license text and full attribution are included in the app's About screen and in every release.