Deckard v0.6.4

Deckard now installs to ~/Deckard instead of the hidden ~/Library/Application Support/Deckard folder. In Chrome's Load unpacked picker, select Deckard > extension in your home folder.

Custom --home locations remain supported. Includes the model-reusing app updates and extension fixes from previous releases.

Install

curl --proto '=https' --tlsv1.2 -fsSL https://github.com/sgoedecke/deckard/releases/download/v0.6.4/install.sh | bash

Open chrome://extensions, enable Developer mode, then choose Load unpacked and select ~/Deckard/extension.

Moving an existing installation

The installer does not automatically relocate an existing installation. If Deckard is installed under Library, first run:

"$HOME/Library/Application Support/Deckard/current/bin/deckard" uninstall

Remove the old Deckard extension in Chrome, run the new installer above, and load ~/Deckard/extension. This fresh installation downloads the full model bundle. Do not move the old folder manually: the native-host registration and shell PATH refer to its installed location.

To upgrade without moving the old installation, retain its location explicitly:

curl --proto '=https' --tlsv1.2 -fsSL https://github.com/sgoedecke/deckard/releases/download/v0.6.4/install.sh | bash -s -- --home "$HOME/Library/Application Support/Deckard"

Then reload Deckard in Chrome and refresh existing web pages. Matching model files are reused with the app-only archive.

SHA256SUMS covers both archives and the installer. Requires Apple Silicon and macOS 15 or newer. Binaries are ad-hoc signed, not notarized.

Deckard v0.6.3

Upgrades no longer download the model again when it is unchanged. The installer verifies the installed model files against the target release checksums and downloads only the 3.2 MB app archive when they match. First installs and missing or changed models use the full bundle.

Also fixes upgrade compatibility with earlier 0.6.x installations. Includes persistent site exclusions and the extension-reload fix from previous releases.

Install or upgrade

curl --proto '=https' --tlsv1.2 -fsSL https://github.com/sgoedecke/deckard/releases/download/v0.6.3/install.sh | bash

Then reload Deckard at chrome://extensions and refresh existing web pages. The installer selects the appropriate archive automatically; no manual model setup is required.

SHA256SUMS covers both archives and the installer. Requires Apple Silicon and macOS 15 or newer.

Deckard v0.6.2

Fixes repeated Extension context invalidated errors after reloading Deckard. Stale content scripts now remove their marks, stop observing the page, and stop retrying an invalid extension connection.

Includes the persistent Do not flag this site checkbox introduced in v0.6.1.

Install or upgrade

Reloading an unpacked extension does not download new releases. Run the installer first:

curl --proto '=https' --tlsv1.2 -fsSL https://github.com/sgoedecke/deckard/releases/download/v0.6.2/install.sh | bash

Then reload Deckard at chrome://extensions and refresh existing web pages once to replace scripts from older versions. The site checkbox appears below the page title in the popup.

Requires Apple Silicon and macOS 15 or newer. SHA256SUMS covers the archive and installer.

Deckard v0.6.1

Adds a persistent Do not flag this site checkbox to the popup. It stops scans and removes flags across tabs for the exact hostname, without affecting subdomains. Preferences survive browser restarts; uncheck the box to resume scanning.

Restores the maintainer-written README, removing recent generated additions.

Install or upgrade

curl --proto '=https' --tlsv1.2 -fsSL https://github.com/sgoedecke/deckard/releases/download/v0.6.1/install.sh | bash

After upgrading, reload Deckard at chrome://extensions. Requires Apple Silicon and macOS 15 or newer. SHA256SUMS covers the archive and installer.

Deckard v0.6.0 — Apple Neural Engine

Default-priority Apple Neural Engine

Deckard now runs Gradient through public Core ML CPU/Neural Engine APIs instead of MLX/Metal. There is no GPU fallback or private ANE runtime hook.

• Includes the pinned FP16 model and tokenizer in the release archive (about 581.5 MiB compressed). No separate Hugging Face download, Python, or Xcode is required.
• Keeps the model resident and caches local compilation across host lifetimes. The first model use can take roughly 26 seconds on an M4; subsequent resident classifications averaged about 188 ms in development measurements. Other Macs may differ.
• Preserves the 50-word minimum, windowing, saved preferences, and default 0.97 cutoff. All 52 frozen reference windows retained the same default-cutoff decisions.
• Supports upgrading existing Deckard MLX installations in place.

Install or upgrade

curl --proto '=https' --tlsv1.2 -fsSL https://github.com/sgoedecke/deckard/releases/download/v0.6.0/install.sh | bash

After upgrading, click Reload on Deckard at chrome://extensions. Apple Silicon and macOS 15 or newer are required. The native executable is ad-hoc signed, not notarized.

The FP16 model is derived from decoded canonical q4 weights; it does not recover the original FP32 weights. Scores remain experimental and are not proof of AI authorship. Model attribution and MIT notices are bundled; see model provenance.

SHA256SUMS covers the archive and installer; the installer also embeds the archive checksum.

Deckard v0.5.0

Adds larger token-aware context windows alongside existing short-passage scanning. Either pass can mark text, with cached threshold updates and overlapping highlights handled together. The default threshold is now 97; saved custom thresholds and On/Off settings are preserved. Native protocol 3 and scanner contract 7 require reloading the unpacked extension and refreshing existing pages after upgrading. Includes owned-installation upgrade support from v0.4.0 and v0.4.1. Apple Silicon macOS 15+; ad-hoc signed, not Developer ID signed or notarized. Detection remains experimental and is not proof of AI authorship.

Deckard v0.4.1

Fixes same-document SPA scanning in Chrome, uses the description 'Local AI-text detection for your browser.', and starts fresh installations On when Chrome grants the declared HTTP/HTTPS website access. Explicitly saved Off settings remain Off. Includes safe in-place upgrades and uninstall ownership for Deckard 0.4.0. Model, threshold and 50-word minimum are unchanged. Apple Silicon macOS 15+; ad-hoc signed, not Developer ID signed or notarized.

Deckard v0.4.0

First Deckard release: local AI-text marking for Chrome using a native quantized Gradient helper. Requires Apple Silicon, macOS 15+, and Chrome 121+. Includes the model, runtime licenses, and fixed-ID unpacked extension. The installer configures PATH and native messaging; Chrome Developer Mode and Load unpacked remain manual. Experimental scores, not proof of authorship. Ad-hoc signed, not notarized. See README for installation and safe uninstall instructions.