Airgap Download

Airgap for macOS, Windows & Linux

Local document AI for confidential files.

Airgap lets you query PDFs, Markdown, and text files with local retrieval, local generation, and citations back to the page. Documents never leave your machine.

Download Watch demo
Download for a specific platform:
Airgap running on a MacBook, answering a question about the Linux VFS with numbered citations beside the cited page of the source PDF.
Formats PDF, Markdown, text
Models Five built in, or any GGUF repo
Citations Page and paragraph
Vaults Portable and shareable

Overview

Document RAG without handing documents to a service.

Airgap builds a private vector index for each vault on your Mac. It parses documents, retrieves the relevant chunks, runs a local language model, and links answers back to the original source pages.

It is designed for local research workflows where precision, provenance, and data residency matter more than cloud convenience.

  1. Someone at a laptop as PDF, Markdown, code, and diagram files gather above it.

    Bring in a project

    Add everything in one place. Docs, code, diagrams, notes, and decisions.

  2. Files, checklists, and conversations connected around a locked safe.

    One project knowledge

    Your team's knowledge becomes a connected, private resource that stays with the project.

  3. A team gathered around a laptop while documents flow toward a new joiner.

    Onboard faster

    New people get up to speed quickly and can ask anything about the project.

  4. An answer on screen with its cited passages highlighted in the source document.

    Trust every answer

    Every answer cites the exact sources with paragraph-level evidence and highlights.

Security model

Built around local custody.

Airgap local custody illustration.

No hosted account

No service account, cloud workspace, or hosted relay sits between the app and your documents.

Local files, disk encryption

Vaults are ordinary files in the app’s own data directory, protected at rest by the disk encryption your machine already has — FileVault, BitLocker, or LUKS. Nothing is uploaded to be indexed.

Offline operation

Ingest, retrieval, and generation continue with Wi-Fi disabled after models are available locally.

Model backends

Use the local model path that fits the Mac.

Airgap model illustration.
Local models Llama 3.2, Qwen 2.5, Phi 3.5, and Gemma 2, run on-device through llama.cpp and cached after one download.
Bring your own Any GGUF model from a Hugging Face repository.
Multilingual retrieval Vaults embed with a 100-language model by default, so documents are searched in the language they are written in.
Connected model Optional and off by default. Point Airgap at a hosted API and only then do questions leave the machine.