Sanctum AI

What is Sanctum AI?

Sanctum AI is a desktop app that runs open-source language models on your own computer, so prompts, answers and documents never need to reach a cloud server. It targets people who want a private chatbot for confidential work.

The app works like a self-contained model manager and chat window. You pick a model from a built-in catalogue drawn from Hugging Face, download it once, and then talk to it or ask it about a PDF while the computer stays offline. During setup you create a password and a recovery phrase, and these protect an encrypted vault that holds your chats and files.

Press coverage names Tyler Ward as a co-founder, and one directory places the company in Brooklyn, New York. Sanctum launched on macOS for Apple silicon and Intel machines, and later reviews describe a Windows build, while Linux has only been mentioned as a plan. The sanctum.ai domain did not resolve during our October 2026 review, so the facts here come from directories, press articles and user write-ups rather than the official site.

Key Features

Local inference inside an encrypted vault is the core of Sanctum AI, and every other feature is built around keeping data on the device.

• Local model runner: Sanctum AI loads open-source models in the GGUF format and runs them on your own processor and memory. Coverage mentions thousands of compatible builds, with Llama, Mistral and Gemma named most often by reviewers.
• Encrypted vault: Chats and uploaded files are stored in a local vault locked by a password and a recovery phrase. One directory reports AES-256 encryption for chat logs and database files, although we could not confirm this on an official page.
• Offline operation: After a model has downloaded, prompts are answered without any internet connection. This makes the app usable on a train, in a secure office or on a laptop that is kept off the network.
• PDF summaries and questions: You can add a PDF and ask the model to summarise it or answer questions about its contents. Early listings called PDF and CSV uploads upcoming, and later reviews describe PDF chat as working.
• Model catalogue and switching: An in-app catalogue lets you browse and download models from Hugging Face, then swap between them in the chat window. Comparing a small 7B model with a larger one on the same prompt takes only a few clicks.
• Templates and resource monitor: Built-in prompt templates cover common jobs such as summaries, while a resource monitor shows how hard the model is working your machine. The monitor helps you judge whether a larger model is realistic on your hardware.

Pros

Confidential files stay on your own disk with Sanctum AI, which is why privacy-conscious professionals choose it over cloud chatbots.

• Data stays local: Nothing has to leave your computer, and reviewers describe a maximum privacy setting with zero data sharing. Lawyers, clinicians and analysts gain a chatbot they can use on material they may not upload.
• Free core app: Directories consistently report a free download covering local models, the vault and offline chat. Your main cost is the hardware you already own rather than a monthly subscription.
• No usage meter: Because inference happens on your machine, there are no per-token API charges or monthly message caps. Heavy users can run long summaries all day without watching a bill grow.
• Wide model choice: One interface gives access to many open-source model families from Hugging Face. You can test which model handles your documents best and switch without installing separate tools.
• Friendly for non-coders: Sanctum wraps downloads, chat and document handling in a consumer-style app with no terminal commands. Reviewers rate the interface as easy to use compared with command-line local AI tools.

Cons

Uncertain availability is the most serious risk with Sanctum AI, because its official domain failed to load during our October 2026 check.

• Unclear status: The sanctum.ai address did not resolve when we tested it, and no official changelog was reachable. Confirm that the download page and updates still work before you rely on the app for client work.
• Hardware demands: Larger models need plenty of memory and disk space, and one directory recommends at least 16GB of RAM. Older laptops may only manage small models, which give weaker answers.
• Answer quality: Small local models still trail leading cloud assistants on reasoning and long documents. One reviewer also found that responses stopped early when the token limit was reached.
• Windows setup friction: A 2025 user write-up described installing a missing DLL by hand before the Windows version would run. Mac users report a smoother start, so test on your own system first.
• No sync or teamwork: Chats live on one device, with no cloud sync, shared workspaces or plugin catalogue. Reports also differ on whether the code is open source, which makes independent auditing harder to judge.

Who is Using Sanctum AI?

Professionals who handle confidential documents make up Sanctum AI’s audience, since they cannot paste client or patient material into public chatbots.

• Legal teams: Solicitors and paralegals use it to summarise contracts and case files offline. Keeping documents local avoids sending privileged material to an outside provider.
• Healthcare staff: Clinicians and researchers use it to draft and summarise notes that contain sensitive details. It supports their work but gives no medical advice, so outputs still need professional review.
• Finance analysts: Analysts summarise internal reports and spreadsheets without sharing them with a cloud service. Any analysis it produces is a drafting aid, not investment advice.
• Developers and researchers: Developers query proprietary code, and academics explore sensitive datasets on their own machines. Educators have also tested it as a private way to try local AI.

Pricing

Sanctum AI is reported as freemium, with a free desktop app and an optional paid tier of about $8 to $10 a month (directory figures, checked October 2026).

Futurepedia lists $8 a month, another directory lists $10, and Vantaige describes the app as entirely free with no paid tier.

Disclaimer: none of these figures came from an official pricing page, so check the price inside the app before you pay anything.

What Makes Sanctum AI Unique?

A password-locked vault with a recovery phrase is the detail that separates Sanctum AI from most local model runners. Tools such as LM Studio and Ollama run models privately, but they generally store chat history as ordinary files, while Sanctum treats encryption at rest as part of the product.

The second difference is the audience. Sanctum packages model downloads, document questions and privacy settings in an app aimed at professionals rather than hobbyists, so a lawyer or analyst can get started without learning about quantisation or command-line flags.

Compatibilities and Integrations

Sanctum AI runs on macOS and, according to later reviews, on Windows, with model downloads coming from Hugging Face.

• Operating systems: macOS on Apple silicon and Intel was the launch platform. Windows is reported by reviewers, and Linux has only been mentioned as a plan.
• Models: GGUF builds from Hugging Face, including Llama, Mistral and Gemma families, are reported. Mistral-7B-Instruct is one model a reviewer tested.
• Files: PDF summaries and questions are reported as working. CSV and Word support appear in some listings, so treat them as unconfirmed.
• Integrations: We found no API, plugin catalogue or Zapier connection. The app is designed to stay self-contained and offline.

Sanctum AI Tutorials

A first private chat takes around 20 minutes, most of it spent downloading the model.

1. Find a working download link for your operating system and install the app.
2. Set a vault password and store the recovery phrase somewhere safe offline.
3. Open the model catalogue and download a small 7B model first.
4. Watch the resource monitor while you ask a few test questions.
5. Add a harmless PDF and ask for a short summary to check quality.
6. Try a larger model only if memory use leaves enough headroom.

How We Rated It

Sanctum AI earns 3.6 out of 5, as excellent on-device privacy and a free core app are offset by thin support and an unreachable official site.

Summary

Sanctum AI suits professionals who want a chatbot for confidential documents without sending anything to a cloud provider. Its encrypted vault, offline operation and free core app make it one of the more privacy-focused local AI options we have reviewed. The main limitation is uncertainty, because the official site did not load in October 2026 and pricing comes only from directories. Answer quality also depends on your hardware, so a modest laptop will limit you to smaller models. Before committing, confirm that a current download exists, then test a small model on a document that contains nothing sensitive.