Pre-alpha — the extension is published, the desktop app builds from source. Don't make it your only safeguard yet.
Roadmap · updated 31 July 2026

Built in the open, phase by phase.

Oath Light is developed as a series of focused phases. The blocking core, the desktop app, the friction system and the recovery layer all work today; Phase 4 is now in verification rather than construction. Here's where it's been and where it's going — no hype, just the plan.

3
phases shipped
4
phase in verification
3
phases planned
v0.5
desktop app, pre-alpha
Phase 1 Shipped

Foundation — the blocking core

The browser extension that does the real work: a curated domain blacklist and a multilingual keyword engine running in a Manifest V3 service worker, with no build step.

385,597 domainsexact + parent matchingadult TLD blocks41-language keywords
Phase 2 Shipped

Desktop companion

A Tauri 2 (Rust) app that keeps protection alive and connects to the extension over a local, authenticated bridge.

Tauri 2 / Rust corenative messagingstartup persistencelocalhost-only
Phase 3 Shipped

Graylist V2 & search enforcement

Per-item NSFW filtering on 35 mixed platforms, permanently forced SafeSearch and YouTube Restricted Mode, and defense against the common bypass tricks.

35 platformsper-item labelsSafeSearch enforcedbypass defense
Phase 4 In verification · you are here

Friction, reach & recovery

Almost everything that can be finished by writing code is finished. What's left needs something else — the real sites open, a GPU for bench runs, someone writing content, or the owner's own release keys. The remaining code items are the installer half of reproducible builds and the string extraction behind translations.

24h delayed weakening dual-process watchdog system DNS filter on-device AI monitor lockdown mode panic flow & streaks AI mentor
Left before Alpha

Real-machine verification of the Edge browser lock and grayscale hours, production OTA keys, a fluent read of the Arabic draft, and a full-scale pre-Alpha launch test.

Needs a GPU

Quantising the model, DirectML/NPU support, a proper 200–400 image eval set, in-page image scoring, and a strictness knob for the AI.

Needs the sites open

Instagram, TikTok, YouTube Shorts, Twitch and Kick — per-item where real labels exist, feed-hiding where they don't. Written blind these would look right and quietly do nothing.

Needs someone who didn't write it

A proper red team. Not gating Alpha, but it should happen before anyone depends on this. Report holes in BYPASSES.md.

Phase 5 Planned

The UI rebuild, then mobile

Finish the design-system rebuild — Blocking, Blocklist and Monitor are still the old screens — then Android: a VPN DNS filter plus accessibility capture, which needs the quantised model first. iOS realistically caps out at Safari rules and a DNS profile.

UI rebuildAndroidon-device AI scanencrypted streak sync
Phase 6 Planned

Reach & polish

Translations beyond the Arabic draft — the locale layer is done, the remaining work is extracting the hardcoded English. Plus reproducible installer builds, performance work, and packaged installers so setup no longer requires building from source.

string extractionmore localespackaged installersreproducible builds
Phase 7 Planned

1.0 & sustained community

A stable, signed 1.0 with the remaining store submissions, and a healthy open-source community keeping the blocklists and code maintained for the long haul.

stable 1.0signed binariesstore submissionscommunity upkeep
Deliberately not doing

So it stops getting re-suggested.

A roadmap is more useful when it also says what's off the table, and why.

Microsoft Edge Add-ons

The listing's verification requirements aren't worth it for a store that wouldn't meaningfully distribute the extension. The consequence is deliberate: Edge can never be force-installed, so a browser lock enforces Edge instead — it stays shut until it's carrying the extension.

Safari, and a mobile extension

Safari has no native messaging, so the desktop bridge can't exist there. Phones get the AI scanner rather than a port of the browser extension.

Score-based blocking

Unpredictable and undebuggable. The filter stays deterministic. Likewise no scraped multi-million domain list — that's how Phase 1 broke the first time.

A 90-day recovery course

The flow runner that would play it stays, but there's no 90-day content coming. Five good exercises beat thirty thin ones.

Roadmaps shift as real feedback comes in. The best way to influence what's next is to try it, file issues, and contribute.