[{"data":1,"prerenderedAt":-1},["ShallowReactive",2],{"project-82770":3},{"id":4,"name":5,"fullName":6,"owner":7,"repo":5,"description":8,"homepage":9,"htmlUrl":9,"language":10,"languages":9,"totalLinesOfCode":9,"stars":11,"forks":12,"watchers":13,"openIssues":14,"contributorsCount":15,"subscribersCount":15,"size":15,"stars1d":16,"stars7d":17,"stars30d":18,"stars90d":15,"forks30d":15,"starsTrendScore":12,"compositeScore":19,"rankGlobal":9,"rankLanguage":9,"license":20,"archived":21,"fork":21,"defaultBranch":22,"hasWiki":21,"hasPages":21,"topics":23,"createdAt":9,"pushedAt":9,"updatedAt":24,"readmeContent":25,"aiSummary":26,"trendingCount":15,"starSnapshotCount":15,"syncStatus":16,"lastSyncTime":27,"discoverSource":28},82770,"pii-guardrail-browser-extension","dfki-dsa\u002Fpii-guardrail-browser-extension","dfki-dsa","Privacy Guardrail is a local-first Chrome extension that detects and anonymizes PII before text is pasted into ChatGPT, Claude, or Gemini, using Rust\u002FWASM recognizers and optional in-browser transformer NER.",null,"TypeScript",71,9,1,13,0,2,14,18,55.8,"Apache License 2.0",false,"main",[],"2026-06-12 04:01:39","\u003Cpicture>\n  \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"docs\u002Fassets\u002Flogo-privacy-guardrail-white.png\">\n  \u003Cimg align=\"left\" alt=\"Privacy Guardrail\" src=\"docs\u002Fassets\u002Flogo-privacy-guardrail-black.png\" height=\"120\">\n\u003C\u002Fpicture>\n\u003Ca href=\"https:\u002F\u002Fwww.dfki.de\u002F\" title=\"Deutsches Forschungszentrum für Künstliche Intelligenz\">\n  \u003Cpicture>\n    \u003Csource media=\"(prefers-color-scheme: dark)\" srcset=\"docs\u002Fassets\u002FDFKI-Logo_ohne_RGB_weiss.png\">\n    \u003Cimg align=\"right\" alt=\"DFKI\" src=\"docs\u002Fassets\u002Fdfki_Logo_digital_black.png\" height=\"72\">\n  \u003C\u002Fpicture>\n\u003C\u002Fa>\n\u003Cbr clear=\"all\">\n\n# Privacy Guardrail\n\nPrivacy Guardrail is a Manifest V3 Chrome extension that detects personally identifiable information (PII) before text is pasted into supported LLM chat apps. Detection runs **entirely on your device**: deterministic recognizers compiled from Rust to WebAssembly, plus optional transformer NER through ONNX Runtime Web. No pasted text leaves the browser, and the project has no telemetry.\n\nDeveloped at the [German Research Center for Artificial Intelligence (DFKI)](https:\u002F\u002Fwww.dfki.de\u002F), Data Science and its Applications research department.\n\n> **Status — public beta (`0.2.0`).** Detection is assistive: it helps you catch personal data before it leaves your machine, but it will not catch everything and is not a compliance or DLP product. See [Known limitations](#known-limitations).\n\n## Contents\n\n- [Supported chat apps](#supported-chat-apps)\n- [Install](#install)\n- [System requirements](#system-requirements)\n- [How it works](#how-it-works)\n- [Known limitations](#known-limitations)\n- [Documentation](#documentation)\n- [For developers](#for-developers)\n- [Roadmap](#roadmap)\n- [Acknowledgements](#acknowledgements)\n\n## Supported chat apps\n\n- ChatGPT (`chat.openai.com`, `chatgpt.com`)\n- Claude (`claude.ai`)\n- Gemini (`gemini.google.com`)\n\nGeneric or custom sites are not supported in this beta.\n\n## Install\n\nEnd users should install from the **Chrome Web Store** once the listing is live. Each GitHub Release also attaches the packaged ZIP and SHA-256 checksum for transparency and manual loading.\n\nFor an unpacked developer install, see [`docs\u002Fdeveloper\u002Fbuilding.md`](docs\u002Fdeveloper\u002Fbuilding.md).\n\n## System requirements\n\n- Chrome desktop stable (latest).\n- **Recommended:** ≥ 16 GB RAM and a WebGPU-capable GPU for smooth Local AI detection.\n- **Minimum for Local AI:** more than 8 GB browser-reported memory. On 8 GB or less, the extension auto-disables Local AI and runs pattern-only detection. Between 8 GB and 14 GB, Local AI stays on but a slowdown warning may appear.\n- Without WebGPU, Local AI falls back to CPU\u002FWASM execution (slower but functional).\n- Pattern-only detection runs on any supported Chrome system regardless of memory or WebGPU.\n\nThese requirements are high because Local AI runs a transformer NER model entirely in the browser. Lowering them is an active roadmap item.\n\n## How it works\n\n- Intercepts text paste events in supported chat inputs.\n- Detects regex\u002Fchecksum-backed PII such as email addresses, phone numbers, SSNs, credit cards, IBANs, IP addresses, and dates.\n- Adds local transformer NER for names, addresses, identifiers, credentials, and other free-text PII when model assets are prepared.\n- Shows a review UI before anonymizing detected spans.\n- Replaces selected spans with stable placeholders such as `[EMAIL_1]` or `[PERSON_1]`.\n- Stores the placeholder map locally in Chrome storage so model responses can be de-anonymized later, with restored values visually highlighted.\n\nNo pasted text is sent to a remote inference service. There is no telemetry or analytics. See [`PRIVACY.md`](PRIVACY.md) for the full privacy posture.\n\n## Known limitations\n\n- Detection can miss sensitive content and can flag harmless text.\n- Short names, ambiguous words, code blocks, tables, and unusual formatting reduce detection quality.\n- Local AI can be slow or unavailable depending on browser, device memory, and WebGPU support; pattern-only mode covers a narrower set of categories.\n- Restoration of placeholders into model responses depends on local records and may not handle every response rewrite.\n\n## Documentation\n\n### For End-Users\n\n- [User guide](docs\u002Fuser\u002F) — install, day-to-day use, Local AI explained, managing local data, troubleshooting, reporting issues safely, detected categories and limitations.\n- [Privacy posture](PRIVACY.md)\n- [Security reporting](SECURITY.md)\n- [Support](SUPPORT.md)\n- [Changelog](CHANGELOG.md)\n- [Impressum \u002F Legal notice](IMPRESSUM.md)\n\n### Project\n\n- [Contributing](CONTRIBUTING.md)\n- [Building from source](docs\u002Fdeveloper\u002Fbuilding.md)\n- [Model assets](docs\u002Fdeveloper\u002Fmodel-assets.md)\n- [Releasing](docs\u002Fdeveloper\u002Freleasing.md)\n\n### For Developers\n\nQuickstart for working on the extension:\n\n```bash\nrustup update\nrustup target add wasm32-unknown-unknown\ncargo install wasm-bindgen-cli --version 0.2.118\nnpm install\n\nnpm run build:wasm\nnpm run dev\n```\n\nLoad `dist\u002F` as an unpacked extension in `chrome:\u002F\u002Fextensions` (Developer mode).\n\nModel-free pull-request checks:\n\n```bash\nnpm run validate:ci\n```\n\nThe full transformer build requires preparing BardsAI EU multilingual NER assets — see [`docs\u002Fdeveloper\u002Fmodel-assets.md`](docs\u002Fdeveloper\u002Fmodel-assets.md). Once prepared, build with strict enforcement:\n\n```bash\nNER_MODEL_ASSETS_REQUIRED=1 npm run build\n```\n\n### Repository layout\n\n- `src\u002F` — extension TypeScript, UI, offscreen detection, benchmark harness.\n- `crate\u002F` — Rust\u002FWASM detection engine.\n- `scripts\u002F` — model prep, packaging checks, benchmark helpers.\n- `benchmarks\u002F` — benchmark harness; generated corpora and reports stay local.\n- `docs\u002F` — user, developer, and design documentation.\n\nBuild and local-model artifacts are intentionally not committed: `dist\u002F`, `crate\u002Fpkg\u002F`, `crate\u002Ftarget\u002F`, `generated\u002Fmodels\u002F`, `.model-sources\u002F`, `.venv\u002F`, `.private-docs\u002F`, `tests-local\u002F`, and `benchmarks\u002Fcache\u002F`. See [`docs\u002Frelease\u002Fpublic-source-boundary.md`](docs\u002Frelease\u002Fpublic-source-boundary.md) for the public source boundary.\n\n## Roadmap\n\nDirectional themes — none are commitments, and order may change with evidence and community feedback:\n\n- More reliable local PII detection (smaller models, distillation, fine-tuning, hybrid pipelines).\n- More browser-efficient inference paths for lower-resource devices.\n- Support for additional Chromium-based browsers beyond Chrome desktop stable.\n- Mobile support for AI workflows on smartphones.\n- Support for additional AI chat platforms.\n\n## Acknowledgements\n\n\u003Ctable width=\"100%\">\u003Ctr>\n\u003Ctd align=\"left\" width=\"120\">\u003Ca href=\"https:\u002F\u002Fwww.dfki.de\u002Fweb\u002Fforschung\u002Fforschungsbereiche\u002Fdata-science-und-ihre-anwendungen\" title=\"Data Science and its Applications, DFKI\">\n  \u003Cimg alt=\"DSA — Data Science and its Applications\" src=\"docs\u002Fassets\u002Fdsa-logo.png\" height=\"120\">\n\u003C\u002Fa>\u003C\u002Ftd>\n\u003Ctd>\nPrivacy Guardrail is developed in the \u003Ca href=\"https:\u002F\u002Fdsa.dfki.de\">Data Science and its Applications research department\u003C\u002Fa> at the \u003Ca href=\"https:\u002F\u002Fwww.dfki.de\u002F\">German Research Center for Artificial Intelligence (DFKI)\u003C\u002Fa>.\u003C\u002Ftd>\u003C\u002Ftr>\u003C\u002Ftable>\n\n### Contributors\n\n- Björn Busch-Geertsema — Lead Developer\n- Sergey Redyuk — Developer\n- Prof. Dr. Sebastian Vollmer — Principal Investigator & Project Originator\n- Rahul Sharma\n- Islam Mesabah\n- Kai Spriestersbach\n- Andrea Sipka\n","Privacy Guardrail 是一个本地优先的 Chrome 扩展程序，旨在检测并匿名化在粘贴到 ChatGPT、Claude 或 Gemini 等聊天应用中的个人身份信息（PII）。其核心功能包括使用 Rust 编译至 WebAssembly 的识别器以及可选的浏览器内基于 ONNX Runtime Web 的转换器命名实体识别（NER），确保所有处理都在用户设备上完成，不会将任何文本发送到外部服务器。适用于需要保护敏感信息不被意外泄露给第三方 AI 服务的场景，如企业内部沟通或个人隐私保护。该扩展由德国人工智能研究中心 (DFKI) 开发，目前处于公开测试阶段 (`0.2.0`)。","2026-06-11 04:09:11","CREATED_QUERY"]