Repository navigation
Installation
DiffLens ships as unsigned-but-verifiable installers built entirely in GitHub Actions from clean checkouts. Grab the artifacts for your platform from the releases page. There is no account, no license key, no server — nothing to configure before your first diff.
| Requirement | Minimum | Notes |
|---|---|---|
| OS | Windows 10/11 x64, macOS 14+ (Apple Silicon), Linux x64 | Intel Macs are not supported in v1 |
| RAM | 8 GB recommended | The e5 embedding model runs comfortably in ~500 MB. The optional Qwen3-4B explanation model wants ≥ 8 GB RAM; the Qwen2.5-3B fallback fits in 6 GB. |
| Disk | ~300 MB for the app | Plus ~118 MB for the required embedding model, ~65 MB for llama-server, and 2.0–2.5 GB for an optional GGUF explanation model (your choice). |
| Display | Any desktop environment | A standard Tauri 2 (WebKit/WebView2/WebKitGTK) window. |
- Download the
.msi(or the NSIS.exe) from the latest release. - SmartScreen shows an unsigned-installer warning because v1 has no Authenticode certificate. Click More info → Run anyway.
- Follow the installer. The
.msiinstalls per-machine; the NSIS.exeis the same app with a wizard UI.
- Download the
.dmg(aarch64) and open it. - Drag DiffLens to
Applicationsand launch it. Because the build is unsigned (no Apple Developer ID for v1), Gatekeeper blocks the first attempt. - Right-click the app → Open, then confirm in the dialog. This is needed once; afterwards macOS remembers your choice.
- Intel Macs: not supported in v1 — there is no x86_64 macOS build.
Two artifacts are published:
-
.AppImage— make it executable first, then run it. It needs FUSE (on Debian/Ubuntu:sudo apt-get install libfuse2).chmod +x DiffLens_1.0.1_amd64.AppImage ./DiffLens_1.0.1_amd64.AppImage
-
.deb— installs system-wide:sudo dpkg -i difflens_1.0.1_amd64.deb # or: sudo apt install ./difflens_1.0.1_amd64.deb
(RPM builds are not published in v1; AppImage or extract-the-deb are the options on Fedora/openSUSE.)
- Every installer is built by the public release workflow on GitHub-hosted runners from a clean checkout — the full build log is public.
- All third-party GitHub Actions in that workflow are pinned to commit SHAs (supply-chain hardening added in v1.0.1).
- Every release publishes SHA-256 checksums, so you can prove the bytes you downloaded are the bytes the workflow produced (next section).
Each release page lists a checksums.txt (or per-artifact hashes in the release notes) produced during the build.
# Option A — one-shot check (file names in checksums.txt must match your downloads)
sha256sum -c checksums.txt
# Option B — manual compare
sha256sum DiffLens_1.0.1_amd64.AppImage
# compare the output to the published hash:
grep AppImage checksums.txtOn macOS use shasum -a 256 <file> in place of sha256sum.
# One file
Get-FileHash .\DiffLens_1.0.1_x64-setup.exe -Algorithm SHA256
# Compare against checksums.txt line by line
Get-Content checksums.txt | ForEach-Object {
$expected, $name = ($_ -split '\s+', 2) | ForEach-Object { $_.Trim() }
$actual = (Get-FileHash $name -Algorithm SHA256).Hash.ToLower()
if ($actual -ne $expected.ToLower()) { "MISMATCH: $name" } else { "OK: $name" }
}A hash mismatch means the download was corrupted or tampered with — delete it and re-download; if it persists, open an issue.
The installer is deliberately small. On first use DiffLens asks to download, through its in-app Downloads screen, the components it needs — each SHA-256-verified against a manifest pinned at compile time:
| Component | Size | Required? | License |
|---|---|---|---|
| multilingual-e5-small (ONNX, quantized) | ~118 MB | Yes (for semantic alignment) | MIT |
llama-server sidecar (llama.cpp b12503) |
~13–25 MB per OS | Only for AI explanations | MIT |
| Qwen3-4B-Instruct-2507 GGUF (Q4_K_M) | ~2.4 GB | No — primary explanation model | Apache-2.0 |
| Qwen2.5-3B-Instruct GGUF (Q4_K_M) | ~2.0 GB | No — low-RAM fallback | Apache-2.0 |
Downloads are resumable (.part files) and cancellable. Details: AI-Explanations-Setup. What is not needed: no account, no telemetry opt-in, no API keys from any cloud provider, no server.