Populyar: Məlumat yeniləndi: 11 noyabr 2026

Kateqoriyalar

İnternet1.405
Ünsiyyət327
Audio və video929
Qrafika və dizayn563
Ofis və məhsuldarlıq1.164
Tərtibatçı alətləri3.260
Süni intellekt alətləri596
Təhlükəsizlik və məxfilik405
Sistem alətləri3.170
Oyunlar542
Təhsil və elm240
Şəbəkə və İT292

Data Maskit köhnə versiyalar və bütün versiyalar

专为大模型打造的本地隐私脱敏与还原网关。

Burada Data Maskit proqramının 11 sentyabr 2026 tarixindən bəri buraxılan 18 versiyası var; son versiya: 0.16.1. Hər versiya naşirin öz quraşdırma faylına aparır.

2026: 18 18 26
VersiyaTarixÖlçüFayllar
0.16.1Son versiya 10 noyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 以前漏发出去的内容,下一轮会补上打码:已确认敏感的值,即使在更早的对话里没被认出来,也会在历史重发时遮住。
- Content that slipped through earlier is now masked on the next turn: confirmed values get masked even in older parts of the conversation that missed them.
- 长对话有时候会突然慢上十几二十秒,现在不会再反复重算同一段历史了。
- Long conversations no longer slow down for tens of seconds by re-processing the same history.
- 日志里那条「有残留(已放行)」现在能点开看到残留的是哪个值、属于什么类型、出现了几次。
- The "residual (forwarded)" log entry now shows which value was left over, its label, and how many times it appeared.
修复
- 修复已发送的历史文本沿用旧结果、导致其后才确认的敏感值继续以原文重发的问题:只在文本真的含该值时重算一次,之后恢复字节稳定。
- Fixed already-sent history replaying an old result, which kept resending a value that was only confirmed sensitive later: the text is now recomputed once, only when it really contains that value.
- 修复短名跨请求不一致:同一个名字在更早的消息里裸出现时,以前每个请求都会被当成普通词跳过。
- Fixed short names being inconsistent across requests: a name that appeared bare in earlier messages used to be skipped as an ordinary word in every request.
- 修复两处缓存口径问题:带已知值的文本一直命中不了结果缓存;模型从不可用变为可用后旧结果不再失效。
- Fixed two cache-key issues: text that went through known-value replacement could never hit the result cache, and cached results stayed valid after the model became available.
- 修复长对话偶发整段重扫、单次请求慢到十几二十秒的问题:映射表满时优先淘汰本轮没用到的条目,正在复用的历史不再被顶掉。
- Fixed long conversations occasionally re-scanning the whole history and taking tens of seconds: when the mapping table is full, entries this round did not reuse are evicted first, so history in active use is no longer pushed out.
- 修复事件详情把处置结论显示成英文裸词 flagged 的问题,并补上出口复检的残留明细(标签 / 长度 / 出现次数 / 掩码预览),不再只报结论不报证据。
- Fixed the event detail rendering the bare English word flagged as the verdict, and added the egress re-check residual details (label / length / occurrences / masked preview) instead of a verdict without evidence.
优化
- 缓存里不再保留没有命中任何敏感词的整段文本副本,内存占用与说明口径一致。
- The cache no longer keeps a copy of text that had nothing to mask, so memory use matches what the documentation claims.
- 运行指标新增「待复核的新值」与「定向重算次数」,面板能看出这条兜底是否在干活。
- Runtime metrics now show pending newly-confirmed values and how often a targeted recompute ran.
- 自检页能看出缓存为什么没命中,设置页会显示日志库当前占多大,方便判断该不该缩短保留天数。
- The self-check page now explains why cached entries were missed, and settings show how much space the log database takes, so you can decide whether to shorten retention.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.16.0...v0.16.1
0.16.0 5 noyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 对话里打码过的名字,换个位置再出现也会被打码,不再以原文发出去。
- Names masked earlier in a conversation stay masked wherever they reappear, instead of going out as plaintext.
- 长对话(尤其是编程助手类客户端)每轮不再把整段历史重新打码一遍,回复明显变快,打码与还原结果不变。
- Long conversations with coding agents no longer re-mask the entire history every turn — replies get faster, masking and restoration unchanged.
- 带图片的对话不再因为图片打码失败而整条发不出去,多图请求也放宽了处理时间。
- Conversations with images no longer fail outright when image masking is slow, and multi-image requests get more time.
修复
- 修复同一对话里已打码的值在无姓名语境的位置被当成误报放行、原文随请求发出的问题:已确认敏感的值在其后出现时直接沿用同一占位符,不再逐处重新判断;请求照常发出,不新增任何阻断。
- Fixed session-known values being skipped as false positives when reappearing without name context and leaking as plaintext: confirmed values now reuse the same placeholder everywhere instead of being re-judged per occurrence; requests are never blocked.
- 修复安全审计的探针结果可能被正常流量的记录挤出读取范围、检查结论误显示为无异常的问题。
- Fixed audit probe findings being evicted from the read window by normal-traffic records, which could render a falsely clean result.
- 修复界面一批问题:自动恢复时间显示成 1970 年、统计页把加载失败显示成无数据、复制按钮在部分环境静默失败、扩展地址复制成固定端口;托盘开关代理失败时不再误显示为已切换。
- Fixed several UI issues: auto-recovery time shown as 1970, stats load failures shown as empty data, silent copy failures in some environments, and the extension address copied with a fixed port; the tray no longer shows a switched state when toggling the proxy failed.
- 关于页拉取更新说明增加超时与失败回退,网络不佳时不再长时间空白。
- The About page now times out and falls back when fetching release notes instead of staying blank on poor networks.
- 修复长对话每轮重复处理整段历史、大请求可慢到十几秒的问题:已处理过的历史直接复用上次结果,打码与还原逐字一致。
- Fixed long conversations re-processing their entire history on every request, which could slow a large request to over ten seconds: processed history now reuses its previous result, with masking and restoration byte-identical.
- 修复首页「已阻断」把仅记录的风险事件也算成拦截、数字虚高一个数量级的问题。
- Fixed the home page's blocked count including flagged-only events and overstating blocks by an order of magnitude.
- 修复用 IPv6 回环地址打开控制台时被拒绝的问题。
- Fixed the console rejecting requests when opened through the IPv6 loopback address.
- 修复界面切英文后托盘菜单仍是中文的问题。
- Fixed the tray menu staying in Chinese after switching the interface to English.
- 修复带图片的对话里,图片已经涂掉却整条请求报错失败的问题:现在照常发出,多图请求也放宽了处理时间,不再因为超时被拦。
- Fixed image-bearing conversations failing with an error even though the images had already been covered: the request now goes through, and multi-image requests get more processing time instead of being blocked by a timeout.
- 损坏或不完整的压缩响应不再被部分放行,而是按失败处理并记录原因;同时升级代理内核修复安全漏洞。
- Corrupt or truncated compressed responses now fail outright instead of being partially accepted; the proxy core was upgraded to patch a security issue.
优化
- 运行指标新增缓存命中率,自检与面板能直接看出长对话是否还在重复处理历史。
- Runtime metrics now include the cache hit rate, so self-check and the panel can show whether long conversations are still being re-processed.
- 升级响应解压组件到修复版本,并给诊断日志加上时间,排查时能定位到具体时刻。
- Upgraded the response decompression component to a patched release and added timestamps to diagnostic logs.
- 面板端口设成桌面端不支持的取值时,界面直接说明原因和替代做法,不再只是连不上。
- When the panel port is set to a value the desktop app cannot use, it now explains why and how to work around it instead of just failing to connect.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.15.1...v0.16.0
0.15.1 4 noyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 普通词语不再被当成人名打码:像「简洁」「边界」这样的日常词,模型读到的是原文而不是占位符。
- Ordinary words are no longer masked as person names, so the model reads your text instead of placeholders.
修复
- 修复普通词语被误判成人名并替换成占位符的问题:短词必须处在明确的姓名语境里才打码,数字、英文短语与过长的片段不再当作人名。
- Fixed ordinary words being mistaken for person names. Short candidates now need clear name context, and numbers, English phrases and over-long fragments are no longer treated as names.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.15.0...v0.15.1
0.15.0 4 noyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 日志里能看出「脱敏的是什么」了:每条显示打码后的形态,不再空白或只显示类别。
- 自定义敏感词的分类名会出现在占位符里(如「人名」),模型能分清人名与单位。
- 图片脱敏定位不准时不再卡死对话,深色背景截图正常可用。
- Log details now show what was masked: each entry keeps the masked form instead of a blank row.
- Custom word categories now appear inside placeholders, so the model can tell a person from an organisation.
- Image masking no longer stalls a conversation when text cannot be located; dark-background screenshots work again.
修复
- 修复「已还原」标记误报:最小记录模式下原本什么都没还原,界面却给每一项都打上了已还原。
- 修复最小记录模式下日志明细只剩类别名、看不到脱敏形态的问题(脱敏词排行榜同样只显示问号)。
- 修复事件详情里「已还原 / 未出现 / 状态未知」三种情况被混为一谈:现在如实区分,状态未记录时说明原因。
- 修复日志页被大量「脱敏后仍有原文」提示刷满的问题:含空格的进程名、以连字符开头的命令行参数不再被当成人名或机构名。
- 修复引擎启动瞬间的一批「令牌无效」拒绝:令牌改为原子写入,界面不会再读到空令牌。
- 修复图片脱敏定位不准时整条对话卡死:改为整图遮盖,图片内容零保留,请求正常完成。
- Fixed a false "restored" badge: in minimal-logging mode nothing had been restored, yet every item was marked as restored.
- Fixed log details showing only category names in minimal-logging mode, with the masked form invisible (the masked-word ranking showed question marks too).
- Fixed the event detail view conflating "restored", "not in this response" and "state not recorded"; it now tells them apart and explains when the state was not saved.
- Fixed the log page being flooded with "original text still present after masking" notices: spaced process names and leading-hyphen command-line flags are no longer treated as person or organization names.
- Fixed a burst of "invalid token" rejections right after engine start: the token is now written atomically, so the UI never reads an empty one.
- Fixed whole-conversation stalls when image text could not be precisely located: the image is now fully covered instead, keeping zero pixel content while letting the request complete.
新增
- 自定义敏感词分类名现在可以出现在占位符里(例如人名、密级),中文分类不再被统一压成一个通用标签。存量日志与对话不受影响。
- 新增分类级「凭据类」标记:被标记的分类只保存打码形态与摘要,原文永不落库。用于把密钥放进中文分类的场景。
- Custom word category names can now appear inside placeholders; Chinese categories are no longer collapsed into one generic tag. Existing logs and conversations are unaffected.
- New per-category "credential" mark: marked categories store only the masked form and a digest, never the plaintext.
优化
- 最小记录模式的词级统计改为保存打码形态(如「张*」)而不是问号,排行榜与明细重新有信息量;原文仍然不落盘。
- Minimal-logging word stats now keep the masked form instead of a question mark, so the ranking and details are informative again; plaintext is still never stored.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.14.1...v0.15.0
0.14.1 1 noyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 修复日志页大量「凭据泄漏」误报:讲解配置示例时不再误报警告。
- 中文姓名识别不再只看开头一段:长文档里的人物姓名现在都能识别到。
- 修复几处影响使用的问题:引擎重启后界面空白、OCR 组件装不上、改配置时反复重连。
- Fixed a wave of false "credential leak" warnings on the log page: configuration examples are no longer flagged.
- Chinese name detection no longer stops partway through: names anywhere in a long document are now detected.
- Fixed several usability issues: blank UI after an engine restart, OCR component install failing, and repeated reconnects while editing settings.
修复
- 修复大量「凭据泄漏」误报:讲解配置示例(如 password=changeme)不再被当成真凭据告警,真凭据仍会告警。
- Fixed a wave of false "credential leak" warnings: configuration examples such as password=changeme are no longer flagged, while real credentials still are.
- 中文姓名识别不再只看前 32KB:长正文里的人物姓名现在都能识别到(旧实现超过 32KB 就整段跳过)。
- Chinese name detection no longer stops at the first 32 KB: names anywhere in a long document are now detected (previously anything past that point was skipped).
- 语义识别开启但缺少运行依赖时,设置页会明确告知缺哪个依赖,而不是静默无效果。
- When semantic detection is enabled but a runtime dependency is missing, the settings page now names the missing dependency instead of silently doing nothing.
- OCR 组件的下载地址会正确传递给后台服务,不再出现“点安装却提示未配置地址”。
- The OCR component download address is now correctly passed to the background service, so "install" no longer fails with a missing-address error.
- 引擎重启后界面不再出现大面积的“令牌无效”空白(此前会一直用旧令牌重试)。
- After an engine restart the UI no longer goes blank with token errors (it previously kept retrying with the stale token).
- 改动客户端配置时不再触发连环重启:短时间内的多次改动会合并成一次。
- Editing client settings no longer triggers a chain of restarts: multiple quick edits are merged into one.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.14.0...v0.14.1
0.14.0 1 noyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 新增「图片内嵌文字脱敏」:截图、文档里的图片、以及 PDF 页面先在本机识别并打码,再发给 AI。
- 开关默认关闭,不开时行为与 0.13.0 完全一致;开了但本机识别服务不可用时,内容会被拦住而不是原样发出去。
- 修复一处文字链路的邮箱漏检:半角冒号后紧跟邮箱地址时不再原样发出去。
- Optional on-device OCR masking for images, pictures embedded in Office files, and PDF pages, before anything is sent.
- Off by default; if it is on and the local OCR service is unavailable, content is blocked instead of being sent unmasked.
- Fixed a text-path email miss: an address typed right after a half-width colon is no longer sent in the clear.
新增
- 新增图片多模态脱敏:代理链路的截图、扩展上传的图片、Office 内嵌图片、PDF 页面都先经本机 OCR 识别,命中的文字在图上物理打码后上行(默认关闭,只走识别与打码,不把图转成文本)。
- Optional multimodal masking: screenshots (proxy path), uploads, Office-embedded images and PDF pages are scanned by a local OCR service and sensitive text is physically redacted (off by default; masking only, never image-to-text).
- PDF 逐页扫描:干净页保留矢量原文,只有含敏感文字的页变成位图;页数/耗时超限时直接阻断并提示拆分。
- PDF support: clean pages keep their vector text, only pages with sensitive text become bitmaps; oversized documents are blocked with a hint to split them.
- OCR 支持指向自建网关的视觉模型(ocr_endpoint + ocr_allow_remote + MASKIT_OCR_MODEL / MASKIT_OCR_API_KEY),提示词与重试按实测调优(标记合规率 50% → 3/3)。
- OCR can point at a self-hosted gateway vision model (ocr_endpoint / ocr_allow_remote / MASKIT_OCR_MODEL / MASKIT_OCR_API_KEY); prompt and retry tuned from measurements (tag compliance 50% → 3/3).
- 设置页新增开关;引擎新增 ocr_endpoint / MASKIT_OCR_ENDPOINT(默认只允许本机地址,自建网关需显式打开 ocr_allow_remote)。
- New settings switch; ocr_endpoint / MASKIT_OCR_ENDPOINT default to loopback only, self-hosted gateways require the explicit ocr_allow_remote opt-in.
- OCR 改为本机 det+rec 引擎(零出网、真检测框),坐标不再依赖模型"报数";打码改为实体级精遮并用「裁图复认」保证不漏字。
- OCR now uses a local det+rec engine (no egress, real detection boxes) instead of model-reported coordinates; masking is entity-level and self-verified by re-recognition.
- 新增「OCR 组件」按需下载(可选运行时):主安装包不变,启用时自动下载/校验/自检/回滚。
- New optional OCR runtime downloaded on demand: the main installer stays small, with automatic verify, self-test and rollback.
- OCR 组件下载地址内置官方默认(固定预发布位,不干扰主程序更新),内网/镜像可用环境变量指向自己的清单;配套发布流水线已就位。
- The OCR component now has a built-in official download location (a fixed pre-release, separate from app updates); air-gapped or mirrored setups can point it at their own manifest, and the release pipeline is in place.
- OCR 输出新增「坐标覆盖率」闸门:模型只标了部分行时按「没定位全」处理(阻断或整页遮盖),不拿漏掉的行赌运气。
- New tag-coverage gate: if the OCR model only tags part of the lines, the page is treated as not fully localised and is blocked or fully covered.
修复
- 修复审计在时间预算截断时不留「解析已跳过”痕迹的问题,排障数据不再在这种路径上缺口。
- Fixed audit time-budget truncation not recording the "parse skipped" marker; diagnostic data no longer has a gap on that path.
- 复用表容量提升 4 倍:重度使用(一天几千次请求)下,早前对话的占位符不再因容量淘汰而偶尔还原不了。
- Reuse-table capacity raised 4x: under heavy daily use, placeholders from earlier conversations no longer occasionally fail to restore due to capacity eviction.
- 修复“接管模式”冒烟验收长期误报:接管其实一直生效(连接复用正常),只是子进程日志被缓冲,判据读不到启动行。
- Fixed a long-standing false failure in the takeover smoke test: takeover was working (connection reuse intact), but the child process log line was stuck in stdio buffering.
- 修复高并发图片识别的验收误判:排队等待属于设计内的保护行为,不再计为失败。
- Fixed a concurrency-test false alarm: queue-wait backpressure is by design and no longer counts as a failure.
- 修复邮箱漏检:「邮箱:」半角冒号后紧跟地址时(无空格)此前不脱敏,纯文字请求同样受影响;连接串密码仍由连接串规则保护,不受影响。
- Fixed an email miss: an address typed directly after a half-width colon (no space) was previously not masked, affecting plain text requests too; connection-string passwords remain protected by the connection-string rule.
- 图片打码框增加上下留白,盖住英文降部与高位数字的笔画残边。
- Image redaction boxes now include vertical padding to cover descenders and tall digits.
- 图片打码框增加随行宽缩放的左右留白:中文前缀较窄的字体(如微软雅黑)上,宽行的实体框起点会系统性右偏导致首位数字露边,现已修复。
- Redaction boxes now add width-proportional horizontal padding: on fonts with narrower CJK glyphs (e.g. Microsoft YaHei), entity boxes on wide lines started systematically too far right and exposed the leading 
0.13.0 30 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 全新视觉体验与暗色优先质感:控制台新增趋势走势图,日志表格自适应小屏幕宽度。
- 图片与文件附件不再被误打码:上传图片与文件时不再损坏二进制数据,避免上游拒收请求。
- 还原归因更精准:普通编号与版本号等普通标识符不再被误判为占位符被改写,状态显示更准确。
- Fresh dark-first visual design, a live trend sparkline on the dashboard, and a responsive logs table.
- Images and file attachments are no longer corrupted by text masking, preventing upstream request rejections.
- Plain identifiers and version numbers are no longer mistaken for rewritten placeholders.
新增
- 全新 macOS 风格设计系统与通用模式组件(PageHeader/Section/StatCard/Toolbar/Hint),统一页面排版节奏与交互质感。
- Mac-styled design system with reusable pattern components (PageHeader, Section, StatCard, Toolbar, Hint) for consistent typography and spacing.
- 控制台新增原生趋势走势图,日志表格优化列宽自适应紧凑屏幕。
- Native trend sparkline on Dashboard and optimized Logs table columns for compact screens.
修复
- 还原失败归因不再把普通标识符当占位符:普通编号与版本号裸串不再计入占位符被改写,假阳性归零。
- Restore attribution no longer flags plain identifiers and version tags as rewritten placeholders.
- SECURITY.md 补注:出口复检的拦截事件按判定结果区分放行与真阻断。
- SECURITY.md now documents that egress-recheck block events distinguish forwarded and blocked decisions.
- 自动更新元数据固定优先取 macOS 的 .app.tar.gz,消除目录枚举顺序依赖。
- macOS update metadata now always prefers .app.tar.gz to avoid 404s on client auto-updates.
- 图片与文件附件不再被当文本脱敏:媒体专属键下的内联二进制整块豁免,浏览器扩展链路同步上报豁免计数。
- Binary images and file attachments under media keys are exempted from text masking, and the extension bridge reports counts in sync.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.12.2...v0.13.0
0.12.2 28 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 日志页不再满屏黄色告警:同一个值在别处重复出现时,不再逐条记成残留。
- 中文姓名识别更准:普通词不再被替换成占位符,正文和代码保持原样。
- 长对话里语义识别不再被整段跳过,检测结果更可靠。
- Fewer yellow warnings in the logs: repeated occurrences of the same value are no longer logged one by one as leftovers.
- Chinese names are recognised more accurately: ordinary words are no longer replaced with placeholders.
- Semantic detection is no longer skipped wholesale in long conversations, so results are more reliable.
修复
- 中文人名检测不再把「简洁」「边界」「项目规」这类普通词当成姓名打码:此前它们会把正文换成占位符,模型据此分析或改代码必然出错。
- Chinese name detection no longer masks ordinary words (common two-to-three-character terms) as personal names, which used to rewrite the text the model reads.
- 形如 hugging-face 的全小写英文复合词不再被当成人名打码。
- Lowercase English hyphenated words such as hugging-face are no longer masked as personal names.
- 出口复检不再对同一个长值在别处的重复出现逐条报「有残留」:本轮已判定为敏感值就全文一并遮盖。
- The egress recheck no longer reports a leftover for every repeat of the same long value: once a value is classified as sensitive this turn, every occurrence is masked.
- 修复长会话里语义识别被整段跳过的问题(已知值正向打码后坐标映射失同步),「检测不完整」告警相应减少。
- Fixed semantic detection being skipped entirely in long conversations (the offset map lost sync after known-value masking), which also cuts the "incomplete scan" warnings.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.12.1...v0.12.2
0.12.1 26 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 修复更新到 0.12.0 后日志页几乎每条请求都标「检测不完整」的问题,告警恢复正常。
- 修复部分请求会整条失败、客户端一直重试失败的问题,涉及工具定义、结构化输出与出口复检。
- 网关不再因为「检测到某个值可能没遮住」就拦截请求:现在只在日志里提示,请求照常转发,不会让微信机器人、编辑器助手这类服务整条不可用。
- Fixed nearly every request showing "incomplete scan" after updating to 0.12.0.
- Fixed requests failing outright and the client retrying forever.
- The gateway no longer blocks requests over a possibly unmasked value: it logs a note and forwards them, so chat bots and editor assistants stay online.
修复
- 「检测不完整」告警不再被长对话误触发:中文姓名补漏只扫前 32KB 是有界设计,此前它把整条请求判成降级,于是长对话(每轮重发整段历史)几乎每条都标黄;现在它仍可见,但不再算作检测未完成。
- 工具定义里 type 写成数组(["string","null"])或对象时不再整条报错:此前请求侧会直接失败(503),响应侧会让整段流式还原降级。
- 上游返回的 choices 不是数组时不再中断还原:此前这类畸形响应会让整段回复以未还原形态透传。
- The "incomplete scan" badge is no longer triggered by long conversations: the 32KB cap on Chinese-name detection is intentional, but it used to mark the whole request as degraded, so long conversations (which resend full history every turn) were almost always flagged.
- Tool definitions whose type is an array (["string","null"]) or an object no longer break the request: previously the request failed outright, and responses fell back to passing text through unmasked.
- Responses whose choices is not an array no longer interrupt restoration.
- 出口复检不再把「本来就该放过的写法」当成泄漏:同一个值在别处以更长数字串、带字母边界、结尾多一个点的形式出现时,规则本就主动放过它,此前会被判成泄漏并让整条请求失败(客户端重试也一样失败)。现在复检与扫描规则同口径,真泄漏仍然拦截。
- 出口复检不再被占位符「拼接」骗过:当对话正文里出现「部分被遮」的示例时,此前把占位符删掉会让两侧文字拼回原文,误判成泄漏;现在占位符按分隔符处理,不会拼出原文。
- The egress recheck no longer treats "forms the rules deliberately skip" as leaks: when the same value appears elsewhere as part of a longer digit run, with letter boundaries, or followed by a trailing dot, the rules intentionally skip it — this used to be flagged as a leak and fail the whole request (client retries failed too). The recheck now uses the same criteria as the scanner, while real leaks are still blocked.
- The egress recheck is no longer fooled by placeholder "splicing": when the conversation text contains a "partially masked" example, deleting the placeholder used to glue the surrounding text back into the original value and misreport a leak. Placeholders are now treated as separators, so no such value can be reconstructed.
- 出口复检默认改为「只记录、不拦截」:命中时日志里留下原因与残留明细,请求照常上行。阻断会让整条业务链路不可用,而漏掉的值可以事后从日志里补规则;需要旧行为的用户可显式开启拦截开关。
- 修复部分地址在含全角字符的文本里被误判为需要打码、而在纯英文文本里又不需要,导致同一条请求被反复拦截的问题。
- The egress recheck now records instead of blocking by default: a hit leaves the reason and the leftover details in the log while the request is forwarded. Blocking takes a whole service offline, whereas a missed value can be fixed later from the log; users who want the old behaviour can turn the block switch on explicitly.
- Fixed addresses being masked in text containing full-width characters but not in plain text, which made the same request get blocked repeatedly.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.12.0...v0.12.1
0.12.0 25 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 数据库密码、API 密钥这类配置项现在能正确识别带前缀的写法,以前只认裸关键词。
- 日志页现在能分清「模型没提到敏感值」和「真的没还原」,不再把两种情况都显示成同一个状态。
- 新增「只提醒」「拦截」两种处理方式和值级白名单,误报可以只放行那一个值,不必关掉整条规则。
- Credentials written the way .env files actually look are now detected; previously only the bare keyword was.
- The log page now tells "the model never mentioned the value" apart from "the value was not restored", instead of showing one status for both.
- Added "warn only" and "block" per-rule handling plus a per-value allowlist, so a false positive can be released without disabling the whole rule.
修复
- 凭据识别覆盖 .env 真实写法:DB_PASSWORD / MYSQL_ROOT_PASSWORD / OPENAI_API_KEY / db.password 等带前缀或点号的键名以前全部漏检,现在正常脱敏。
- 工具定义区不再被改写:工具名、命名空间、参数结构、语法约束以前会被打码,导致上游拒绝请求;现在只扫描其中的说明文字。
- 视觉等价写法不再绕过:零宽字符插入、全角数字与符号(138…、alice@example.com)以前整条漏检,现在能识别,且发出去的内容与用户输入逐字节一致。
- 已知敏感值不再重复出网:某个值一旦被确认敏感,之后在任何位置出现都会被替换,不必再靠 password= 这类锚点。
- 新增出口前复检:脱敏后仍能匹配到原文的请求会被拦下,不再上行。
- 新增中文姓名补漏与剔误:客户陈建国 这类句式不再漏检,张开的括号 这类普通词不再被当成人名。
- 还原状态细分:no_placeholder_in_response 拆成「模型没提」「模型提了但形态坏了」「响应没走完」,日志页可区分。
- 「只提醒」「拦截」两种处理方式现在对所有规则都生效(此前只对部分规则有效),且放行记录会显示在日志里。
新增
- 每条规则可选「脱敏 / 只提醒 / 拦截」三种处理方式。
- 值级白名单:可放行单个具体值,大小写不敏感,最多 200 条。
- 脱敏取证:可在日志页查看最近 50 条「实际发出去的内容」,仅内存、不落盘。
- 高熵凭据检测(默认关闭):覆盖没有厂商前缀、也没有赋值标签的内部服务令牌。
Fixed
- Credential detection now covers real .env key names (DB_PASSWORD, MYSQL_ROOT_PASSWORD, OPENAI_API_KEY, db.password), which were previously missed entirely.
- Tool definitions are no longer rewritten: tool names, namespaces, parameter schemas and grammar constraints used to be masked, causing upstream rejections; only their descriptions are scanned now.
- Visually equivalent spellings no longer bypass masking: zero-width insertions and full-width digits/symbols (138…, alice@example.com) were missed entirely; they are now detected, and what leaves the machine stays byte-identical to what the user typed.
- A value once confirmed sensitive is now masked wherever it appears, without needing a password=-style anchor.
- New pre-egress re-check: a request that still matches an original value after masking is rejected instead of being sent upstream.
- Chinese name completion and rejection: "role word + name" phrases (customer, contact, author …) are no longer missed, and ordinary words that merely start with a surname are no longer treated as names.
- Restore status is now split: no_placeholder_in_response becomes "the model never mentioned it", "the model rewrote the placeholder", or "the response did not finish".
- "Warn only" and "block" now apply to every rule (previously only some), and what was let through is shown in the log.
Added
- Each rule can be set to mask, warn only, or block.
- Per-value allowlist: release a single concrete value (case-insensitive, up to 200 entries).
- Masking forensics: the log page can show the last 50 bodies actually sent upstream (memory only, never written to disk).
- High-entropy credential detection (off by default) for internal service tokens with neither a vendor prefix nor an assignment label.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.11.4...v0.12.0
0.11.4 20 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 客户端名称现在支持空格,不再出现输入空格导致配置重置为默认客户端的问题。
- 敏感词库支持一键导出备份与导入,方便跨设备迁移与词条管理。
- 敏感词库增加大小写自动覆盖与中文整词匹配指引,词条匹配更清晰。
- Client names now support spaces without resetting your configured clients to defaults.
- Sensitive words can now be exported and imported for easier backups and multi-device setup.
- Word lists now include case-insensitive matching tips and clearer guidance for Chinese words.
修复
- 客户端名称支持空格:输入包含空格的客户端名称时不再被忽略,更不会因所有客户端失效而误把配置重置为默认客户端。
- 客户端修改校验:保存客户端时对名称与目标地址进行前置检查,非法输入即时提示,避免配置损坏。
新增
- 词库导入导出:设置页敏感词库新增一键导出备份,支持通过文件或粘贴导入,提供增量合并与完全覆盖两种模式。
- 词条变体提示:同分类下录入大小写变体时主动提醒已自动覆盖,并补充中文独立词正则用法说明。
Fixed
- Client names with spaces are now accepted and no longer trigger an accidental reset to default clients.
- Added preflight checks for client name and target URL to catch invalid input before saving.
Added
- Word list export and import: backup custom sensitive words as a JSON file, or import via file or paste with merge and overwrite options.
- Case-variant duplicate hints and clearer guidance on regex patterns for isolated Chinese words.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.11.3...v0.11.4
0.11.3 20 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 流式回复不会再中途断掉:模型最后那段回答以前可能整段丢失,客户端还会误报成网络异常。
- Windows 安装装得更干净:不再弹一段红色报错,桌面上也不会多出一个重复图标。
- 同一局域网的其它设备现在也能打开控制台,需要先在设置里手动开启。
- Streaming replies no longer break near the end — the model's final chunk used to vanish and clients reported a network error.
- The Windows installer is cleaner: no red error output, and no duplicate desktop icon.
- Other devices on your network can now open the console, after you turn it on in settings.
修复
- 流式回复不再整段丢失:响应里出现对象或数组形态的字段时还原会中断,客户端只收到后半截数据。
- Windows 安装与卸载不再弹 PowerShell 报错;那条终止旧进程的命令此前每次都语法失败,进程实际没被杀掉。
- 安装不再多出一个桌面快捷方式:改由安装程序按「创建桌面快捷方式」勾选与升级模式决定。
- 反馈模板里的诊断包入口改回真实路径。
新增
- 局域网访问开关:设置 → 高级选项,开启后同一局域网的其它设备可访问控制台与代理端口(重启引擎生效)。
Fixed
- Streaming replies are no longer truncated when a response field holds an object or array instead of a string.
- The Windows installer and uninstaller no longer throw a PowerShell syntax error; that command had never actually terminated the old processes.
- The installer no longer creates a duplicate desktop shortcut — creation follows the checkbox and the update mode.
- The diagnostics-bundle entry in the bug report template now points at the real path.
Added
- LAN access toggle under Settings → Advanced; other devices on the same network can reach the console and proxy ports (engine restart required).

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.11.2...v0.11.3
0.11.2 19 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
更新摘要 / Summary
- 更新弹窗和「关于」页现在直接告诉你这次改了什么,不再只有一句「修复 2 项」或一整段技术日志。
- Windows 桌面版不会再出现「点了启动却起不来」:端口探测多了一条兜底,不再单靠系统命令。
- 启动失败时不再只有一句「未就绪」:端口被占用、旧引擎残留、令牌没写出来会分开说明,日志也不会再因为进程被杀而整片丢失。
- The update card and About page now tell you what actually changed, instead of a bare count or a wall of technical text.
- Windows desktop no longer fails to start: port detection has a fallback when the system command reports nothing.
- A failed start now says why — port taken, stuck engine, or missing token — and engine logs survive a force-kill.
修复
- Windows 启动:端口探测不再只依赖系统命令,命令被拦截或超时返空时改用连接探测兜底。此前该命令一旦失灵,就绪判定永远认为端口未监听,点启动会等满 60 秒后报失败,常驻守护还会反复重启正常工作的代理。
- 引擎日志:壳层强杀引擎时,engine-stdout.log 的块缓冲来不及 flush,内存里的诊断行随进程一起消失(引擎存活期间该文件恒为空)。现由壳层注入 PYTHONUNBUFFERED,并把 [panel] / [engine] / [proxy] 诊断行落盘到数据目录 engine.log(2MB 轮转);SHIELD 事件行携带原文预览,仍然只进内存。
- 引擎启动:守护线程的假死判定(6 轮 × 2s ≈ 12s)比就绪预算(20s)更早,会把正在初始化的引擎强杀重拉,慢机上表现为「永远起不来 / 反复重启」;现加与就绪预算同值的启动宽限。就绪超时也不再只报一句「未就绪」:区分端口被第三方进程 / 本产品残留引擎占用、系统查不到占用 PID、令牌未写出,并把启动失败现场补写进 crash-dumps。
优化
- 版本说明:更新弹窗与「关于」页只展示「更新摘要」,不再把整章 CHANGELOG(含 ### 标记与双语技术条目)原样渲染给用户;桌面端 latest.json 的 notes 也从「修复 N 项」改为摘要正文。
- 版本说明:新增 scripts/check-changelog.py 门禁(已进 scripts/verify-all.py 的 version 组与 ci.yml),强制每个版本章节开头有中英双语「更新摘要」,并拦截反引号、源码文件名、文件:行号、4xx/5xx 状态码与硬技术术语。历史版本(0.10.1 之前)不回溯。
Fixed
- Windows start: port detection no longer relies solely on the system command; when it is blocked or returns nothing, a connection probe takes over. A failing command previously made the readiness check see no listeners forever — start waited the full 60 s and reported failure, and the resident watchdog kept restarting a healthy proxy.
- Engine log: a force-killed engine never got to flush its block-buffered stdout, so engine-stdout.log stayed empty for the engine's whole lifetime and every diagnostic line died with the process. The shell now injects PYTHONUNBUFFERED and persists [panel] / [engine] / [proxy] lines to engine.log in the data directory (2 MB rotation); SHIELD event lines carry plaintext previews and stay in memory only.
- Engine start: the watchdog's hang detection (6 rounds × 2 s ≈ 12 s) fired before the 20 s readiness budget and killed engines that were still initialising, which on slow machines looked like "never starts / restarts forever"; a start grace period now matches the readiness budget. A readiness timeout also reports why (port held by a third-party process, a stuck engine of ours, an unidentifiable listener, or a token that was never written) and dumps the failure scene into crash-dumps.
Improvements
- Update notes: the update card and About page now render only the user-facing summary instead of the whole CHANGELOG section (raw ### markers and bilingual technical entries); the desktop latest.json notes carry the summary body instead of an "N fixed" count.
- Update notes: new scripts/check-changelog.py gate (added to scripts/verify-all.py's version group and ci.yml) requires a bilingual summary section at the top of every version section and rejects backticks, source filenames, file:line refs, 4xx/5xx status codes and hard jargon. Versions before 0.10.1 are exempt.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.11.1...v0.11.2
0.11.1 16 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
修复
- 引擎:还原失败不再把「本机从未签发过」的占位符算进去(客户端历史携带的外部 token / 对话里的文档示例 / 模型自造),新增「外部占位符」独立计数,面板与详情页分开显示。
- 引擎:语义识别不再把代码标识符(JSON.parse、FSPAGE_9105 这类)判成人名/机构名掩码,模型看到的代码不再被改写(真机当日 NAME 命中里六成是变量名/方法名)。
Fixed
- Engine: restore failures no longer include placeholder-shaped strings this instance never issued (foreign tokens from client history / doc examples / model-invented); a separate "foreign tokens" counter is shown apart from restore failures on the dashboard and in the detail dialog.
- Engine: NER no longer masks code identifiers (JSON.parse, FSPAGE_9105, …) as person/organization names, so the code a model sees is no longer rewritten (six in ten of the day's NAME hits were variable or method names).

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.11.0...v0.11.1
0.11.0 14 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
新增
- 面板日志:新增「结论」列(成功 / 有告警 / 失败 / 已取消 / 进行中 / 未完成 / 未处理 / 安全告警),一眼能看出请求成败;事件类型降为小图标。
- 面板日志:取消与错误事件按会话折进对应的脱敏行,一次请求只占一行,不再只看到半条链路。
优化
- 面板日志:列表由 9 列收到 8 列(入口并入上游列);状态列显示 HTTP 码或失败归因(客户端断开 / 上游静默 / 域名解析失败),不再显示截断的英文状态码。
- 面板日志:耗时列区分端到端与仅脱敏耗时;类型筛选改为「按结论」优先,底层原语收进高级分组;图例改为结论颜色说明。
- 面板日志:显式选择筛选条件时不再被「隐藏噪声」吞掉(此前选「用户取消」只会得到空列表)。
Added
- Panel logs: new Outcome column (success / warnings / failed / canceled / in progress / incomplete / not processed / security alert) so request success is visible at a glance; the event type is now a small icon.
- Panel logs: cancel and error events are folded into their masking row per session, so one request is one row instead of half a chain.
Improvements
- Panel logs: 9 columns reduced to 8 (ingress merged into the upstream column); the status column now shows the HTTP code or the failure attribution (client disconnected / upstream silent / DNS failure) instead of a truncated English status.
- Panel logs: the duration column distinguishes end-to-end from masking-only time; the type filter is now outcome-first with raw primitives tucked into an advanced group; the legend explains outcome colours.
- Panel logs: an explicit filter selection is no longer swallowed by "hide noise" (previously picking "user canceled" returned an empty list).
修复
- 前端:复制回退失败时不再谎报「已复制」,设置页会正确提示复制失败。
- 前端:更新提示的版本号不再重复 v 前缀(Web / Docker 端 GitHub tag 自带 v,与模板叠加后实测显示成 vv0.11.0)。
- 引擎:还原详情区分「模型回显但未还原」与「模型未回显」,不再把后者渲染成「未还原」告警;未还原计数不再被日期串等正常文本误触发;签名/密文思考块的契约豁免在还原侧也计数可见。
Fixed
- Frontend: a failed clipboard fallback no longer reports "copied"; the settings page now shows a copy-failed toast.
- Frontend: the update prompt no longer doubles the v prefix on the version (on Web/Docker the GitHub tag already carries v, which stacked with the template and rendered vv0.11.0).
- Engine: restore details now distinguish "echoed but not restored" from "not echoed"; unresolved counting no longer fires on ordinary text such as date strings; restore-side contract skips of signed/encrypted thinking blocks are counted and visible.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.10.1...v0.11.0
0.10.1 13 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
修复
- 引擎:上游发完 [DONE] 后未按规范收尾响应体时,流式请求不再丢失还原记录,正常结束也不会被误记为「客户端取消」。
- 引擎:上游 target 变更后,接管连接池按新目标白名单自动换代,不再对配置内的合法目标返回 403。
Fixed
- Engine: when an upstream omits the closing chunk after [DONE], streaming requests no longer lose their restore record or get misreported as a client cancel.
- Engine: after an upstream target change, the takeover pool now rebuilds against the new target allowlist instead of rejecting configured targets with 403.

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.10.0...v0.10.1
0.10.0 11 oktyabr 2026 — 64 bit EXE (NSIS)
Versiya qeydləri
新增
- 统计:支持各大模型 Token 缓存采集与日摘要聚合,费用估算按厂商口径支持缓存折扣。
Added
- Stats: support token cache extraction and daily aggregation, with vendor-specific discount support in cost estimation.
优化
- 引擎:接管连接池空闲保留由 30 秒提到 120 秒,建连超时放宽至 20 秒并细化 502/504 错误归因,减少冷握手超时与假报错。
- 脱敏:默认凭据前缀安全扩充常见供应商识别(仅对未自行配置过的用户生效,已自定义保存过的请手动添加)。
Improvements
- Engine: takeover pool keepalive budget extended to 120s, connect timeout relaxed to 20s with refined 502/504 attribution to reduce false connect failures.
- Masking: expanded default credential prefixes for common providers (applies only when not customized; manual addition needed for existing custom configs).

扩展更新提示 / Extension update notice
✅ 本次发版扩展无改动:已安装的扩展无需重新加载。
✅ The browser extension did not change: no need to reload it.
Full Changelog: https://github.com/xiaYuTian11/maskit/compare/v0.9.0...v0.10.0
0.2.7 11 sentyabr 2026 — 64 bit EXE (NSIS)

Sİ köməkçiləri və LLM: ən populyar

Hamısına bax
  1. Claude CodeSİ köməkçiləri və LLM · Anthropic PBC
  2. Codex CLISİ köməkçiləri və LLM · OpenAI, Inc.
  3. ChatGPT by lencxSİ köməkçiləri və LLM · lencx
  4. GPT4AllSİ köməkçiləri və LLM · Nomic
  5. JanSİ köməkçiləri və LLM · Jan
  6. oh-my-piSİ köməkçiləri və LLM · Can Bölük
  7. T3 CodeSİ köməkçiləri və LLM · T3 Tools
  8. WSL ManagerSİ köməkçiləri və LLM · Bostrot
  9. CC SwitchSİ köməkçiləri və LLM · farion1231
  10. LobeHubSİ köməkçiləri və LLM · LobeHub

Teqlər

Paylaş