OpenMinis

Your private, on-device AI agent.

OpenMinis brings leading models — Claude, GPT, Gemini and more — into a native mobile experience, and gives them a real computer to work with: a full Linux shell running on your device, browser automation, extensible skills, persistent memory, and deep system integration.

It is free, and fully open source.

We believe that in the age of AI, technical design and code are no longer where a product's advantage lies. The best agent emerges from a tight feedback loop with the people who use it — their expectations and their reports are what converge on the product.

Official website: openminis.app

 

---

What it does

---

What you can do with Minis

A few things people actually use it for:

• Photograph a meal, log the nutrition — Minis identifies the dishes, estimates calories and macros, and writes them to Apple Health.
• Wake up to your timeline — Shortcuts triggers Minis to fetch your X timeline, summarise it, synthesise speech, and play it as your alarm.
• Turn group chatter into tasks — pull messages from a Telegram group, extract bugs and action items, deduplicate them, and file them into Apple Reminders.
• Mount your Obsidian vault — research, clean up and write Markdown notes back into the vault as a normal workspace.
• Share anything into a calendar event — send a page or message to Minis via the iOS Share Sheet and it creates the event, time and place included.

→ OpenMinis/AwesomeMinis — a curated, community-contributed collection of use cases and workflows across health, productivity, research, finance and developer tooling.

---

Skills

A skill is a folder with a SKILL.md file — instructions, and optionally scripts, references and assets — that the agent loads on demand when a request matches it. Metadata stays in context for triggering; the body and bundled resources load only when the skill is actually used.

Minis has its own tool system, but it does not require skills written specifically for it: skills built for Claude, Codex, OpenClaw or Hermes Agent generally run in Minis as-is. Skills that have been adapted to Minis' tools simply run better — they can reach the Linux shell, device integrations and native offloads directly.

→ OpenMinis/MinisSkills — skills adapted for Minis alongside ones built for it from scratch, covering TTS, search, media downloads, health analysis, cloud APIs and more.

---

Press

"the most impressive indie app I've seen in a while"

— Federico Viticci, Open Minis Is the iOS Agent I Wish Siri AI Could Be, MacStories (July 2026)

"在很大程度上实现甚至局部超越了 Apple Intelligence"

— Ye Han, 这可能是 iPhone 最强 Agent 软件，没有之一 丨Open Minis 入门指南, 知乎 / Zhihu (June 2026)

"可能是 iOS 端最强 AI Agent"

— Open Minis：可能是 iOS 端最强 AI Agent, 小众软件 / Appinn (March 2026)

---

Beta programme

App Store releases can lag behind: every update waits on review, and we hold builds back when stability warrants it. The TestFlight build is where fixes and new features land first.

→ Join the TestFlight beta

On Android, the releases page always carries the latest APK.

---

Building from source

Minis ships a Linux sandbox inside the app, so the native dependencies (iSH on iOS, PRoot on Android, FFmpeg, LAME, rclone) and the Alpine rootfs are built from source rather than committed as binaries.

→ See BUILDING.md for the full first-build guide.

The short version:

git clone --recurse-submodules https://github.com/OpenMinis/OpenMinis.git
cd OpenMinis

# iOS  — needs Go 1.25+ and Xcode's Metal Toolchain
#        (xcodebuild -downloadComponent MetalToolchain)
#        order matters: FFmpeg links against LAME
./deps/build_lame.sh && ./deps/build_ffmpeg.sh
./deps/build_ish.sh && ./deps/prepare_alpine_rootfs.sh
./deps/build_rclone_ios.sh
open src/ios/Minis.xcodeproj

# Android — needs NDK r28+, Go 1.25+ and gomobile
./deps/build_proot.sh && ./scripts/prepare_android_sandbox.sh
./deps/build_rclone_android.sh
mkdir -p src/android/app/libs && cp deps/build/rclone/rclone.aar src/android/app/libs/
cd src/android && ./gradlew :app:assembleDebug

BUILDING.md covers the toolchain requirements per platform, the build-time customization templates, and a troubleshooting section for the failure modes you are most likely to hit.

---

Repository layout

src/ios/          iOS app (Swift / SwiftUI) + share, widget and file-provider extensions
src/android/      Android app (Kotlin / Compose) + JNI native code
src/shared/       Assets shared by both platforms
deps/             Native dependency build scripts and vendored sources
docs/specs/       Architecture and interface specifications
scripts/          Rootfs preparation and developer tooling

---

Acknowledgements

OpenMinis stands on a great deal of open-source work. Our thanks to the maintainers of these projects — the full inventory, with versions and license terms, is in THIRD_PARTY_LICENSES.md.

The sandbox — the heart of the product:

• iSH (GPLv3) — Linux usermode emulation on iOS. We run an ARM64 fork.
• PRoot (GPLv2) — user-space chroot for the Android sandbox, via our fork; talloc (LGPLv3+) underpins it.
• Alpine Linux — the minirootfs the sandbox boots.

Media & text — FFmpeg (LGPL-2.1+), LAME (LGPL), cppjieba (MIT), KaTeX (MIT).

iOS — SwiftAnthropic, SwiftMath, RealTimeCutVADLibrary (all MIT), swift-cmark (BSD-2-Clause), and the Apple / Swift Server Workgroup packages (Apache-2.0).

Android — AndroidX & Jetpack Compose, OkHttp, Coil, kotlinx serialization & coroutines, multiplatform-markdown-renderer, Reorderable, ACRA (all Apache-2.0), and Shizuku (MIT).

---

License

OpenMinis is licensed under the GNU General Public License v3.0.

The app links GPL-licensed components — iSH (GPLv3) and PRoot (GPLv2) — so the combined work is distributed under GPLv3. Bundled third-party licenses are listed in THIRD_PARTY_LICENSES.md.

---

Community

• Telegram: Join the group
• Issues: Bug reports, feature requests and discussion via GitHub Issues

This repository is a mirror of a private development tree, so it does not accept pull requests — there is nowhere for them to land. Issues are the way to shape the product, and AwesomeMinis and MinisSkills both do take contributions. See CONTRIBUTING.md.

Highlights / 重点

• Sub Agents — delegate work to background agents you define, each with its own model and reasoning level; steer, stop or resume them from their card. 子代理 —— 把任务交给自定义的后台代理，每个代理有自己的模型和推理强度；可在卡片上纠正、停止或恢复。
• Backup & Restore rebuilt — run history, background backups, local-folder destination, and packages that restore across iPhone and Android in both directions. 备份与恢复重做 —— 备份记录、后台备份、本地文件夹目标，备份包可在 iPhone 与 Android 之间双向恢复。
• Tablet & desktop-class input — two-pane layout with a draggable divider, hardware-keyboard shortcuts and mouse text selection. 平板与键鼠 —— 双栏布局（分隔线可拖动）、硬件键盘快捷键、鼠标拖选文字。
• 11 new languages — Spanish, Indonesian, Malay, Filipino, Thai, Turkish, Polish, Romanian, Brazilian Portuguese, Croatian and Traditional Chinese; existing languages are now fully translated. 新增 11 种语言 —— 西班牙语、印尼语、马来语、菲律宾语、泰语、土耳其语、波兰语、罗马尼亚语、巴西葡萄牙语、克罗地亚语和繁体中文；原有语言补齐全部翻译。
• GitHub Copilot, Claude Opus 5.5 / Sonnet 5.5 / Fable 5.1, GPT-6 and custom per-model request settings. GitHub Copilot、Claude Opus 5.5 / Sonnet 5.5 / Fable 5.1、GPT-6，以及按模型自定义请求参数。

New features / 新功能

Agents / 代理

• Sub Agents: define agents in Settings › Sub Agents with a pinned model and an optional reasoning level; the model delegates to them with one tool. 子代理：在「设置 › 子代理」中定义代理，可固定模型并单独设置推理强度；模型通过一个工具即可委派任务。
• Delegations run in the background and report back with a callback card; past the concurrency limit they queue (up to 10) instead of being refused. 委派任务在后台运行，完成后以回调卡片汇报；超过并发上限时排队（最多 10 个），不会被拒绝。
• A running sub agent can be corrected mid-task, stopped from its card, and an interrupted one resumed. 运行中的子代理可随时纠正方向、从卡片停止，中断的可一键恢复。
• Sub agent cards show the model and reasoning level that actually ran, the live tool, and a readable error when one fails. 子代理卡片显示实际使用的模型与推理强度、正在执行的工具，失败时说明原因。
• Per-tool switches for the browser and delegation under Settings › Agent Runtime › Tools. 「设置 › Agent 运行时 › 工具」中可单独开关浏览器与委派工具。
• A turn's tool calls now run concurrently, and shell commands run in parallel across chats. 同一轮的多个工具调用并发执行，不同会话的 shell 命令也可并行运行。

Chat / 聊天

• Switching model mid-task takes effect from the next request, not after the whole turn. 任务进行中切换模型，从下一次请求起生效，不必等整轮结束。
• Long pasted text folds into a [Pasted#N] chip; tap it to see the full text. 长文本粘贴折叠为 [Pasted#N] 标签，点击查看全文。
• New message actions: Delete From Here, Compact Above, Copy Full Text / Plain Text, Read Selection / Read from Start. 新增消息操作：从此处删除、压缩以上内容、复制全文 / 纯文本、朗读选中 / 从头朗读。
• Per-message token usage (tap the blank area under a reply), including cache hit rate and output speed. 每条回复的 Token 用量（点击回复下方空白处显示），含缓存命中率和输出速度。
• The composer glows as the context window fills up, and shows the size after compaction. 上下文接近上限时输入框会发光提示，压缩后显示压缩后的大小。
• Long-press Send to choose what Return does; the input placeholder rotates through feature tips. 长按发送按钮可设置回车键行为；输入框提示语轮播功能提示。
• Tables keep their key columns on screen, follow markdown alignment and size columns by content; long formulas wrap instead of being clipped. 表格优先保留前几列、遵循 Markdown 对齐方式并按内容分配列宽；长公式自动换行，不再被截断。
• minis:// folder links open the in-app file browser; long-press the tool pager arrows to jump to the first or last call. minis:// 文件夹链接直接打开应用内文件浏览器；长按工具翻页箭头可跳到第一个或最后一个调用。
• "New Chat" in the Move-to picker, with the most recently active group listed first; multi-select Export on the session list. 「移动到」可直接新建会话，最近活跃的分组排在最前；会话列表多选后可导出。

Scheduled tasks / 定时任务

• Fired tasks appear as cards with a dedicated detail page (status, progress, parameters, every fire), and can be cancelled from the card. 定时任务触发后显示为卡片，有独立详情页（状态、进度、参数、每次触发记录），可在卡片上取消。
• A task can run a fixed shell command before asking the model, and can carry a thinking level. 任务可在询问模型前先执行固定的 shell 命令，也可指定思考强度。
• A fire in a busy chat slips in at the next tool boundary and the interrupted work then resumes; tasks created in a chat fire back into it. 会话忙碌时，定时触发会在下一个工具间隙插入，之后恢复原任务；在会话中创建的任务默认回到该会话触发。
• minis-scheduled gains --after, --interval/--count, on-completion chaining and --thinking. minis-scheduled 新增 --after、--interval/--count、完成后串联和 --thinking 参数。

Providers & models / 服务商与模型

• GitHub Copilot sign-in (device flow), with a consent notice before it starts. 支持 GitHub Copilot 登录（设备授权流程），开始前会显示授权说明。
• New models: Claude Opus 5.5, Sonnet 5.5 and Fable 5.1; GPT-6 Sol, Luna and Astra (Codex); GPT Image 2.5. Codex OAuth now discovers its models automatically. 新模型：Claude Opus 5.5、Sonnet 5.5、Fable 5.1；GPT-6 Sol、Luna、Astra（Codex）；GPT Image 2.5。Codex OAuth 自动发现可用模型。
• Per-model overrides: temperature, top-p, custom headers and extra request body fields. 按模型自定义：temperature、top-p、自定义请求头和额外请求体字段。
• xAI Fast Mode (Priority Processing) per provider; OpenCode Go works. xAI 按服务商开启 Fast Mode（优先处理）；支持 OpenCode Go。
• Model lists refresh every 6 hours and on return to the app; model search stays fast with thousands of models; each picker section links to its provider's settings. 模型列表每 6 小时及回到应用时刷新；上千个模型下搜索依然流畅；选择器中每个服务商可直接跳到其设置。

Voice / 语音

• Send waits for speech recognition to finish; a failed recognition offers a retry instead of dropping the audio. 发送会等语音识别完成；识别失败时提示重试，不再丢弃录音。
• After 5 s without a result, a "Switch Model" pill retries on the next voice model. 5 秒无识别结果时出现「切换模型」按钮，可换下一个语音模型重试。
• AI correction uses what is on screen and restores English terms; speech follows the app language. AI 纠错会参考屏幕内容并还原英文术语；朗读语言跟随应用语言。
• A microphone held by another app (a call, a recorder) is now reported clearly. 麦克风被其他应用（通话、录音）占用时会明确提示。

Backup & restore / 备份与恢复

• Backup history with a detail page per run, a running view with Stop, and backups that keep going in the background. 每次备份都有记录和详情页，运行中可查看进度并停止，退到后台也会继续。
• Local folder as a destination; back up without any destination and save or share the package; name your device. 支持本地文件夹作为备份目标；无目标时也可备份并保存或分享备份包；可自定义设备名。
• Restore: browse destinations folder by folder, swipe to delete a package, progress with ETA and a working Stop. 恢复：逐级浏览备份目标、左滑删除备份包，显示进度和剩余时间，可随时停止。
• Delete a backup's packages from the servers too; the local package is removed once every destination is verified. 可同时删除服务器上的备份包；所有目标验证成功后自动清理本地备份包。
• "Have a backup? Restore from it" during first-time setup. 首次设置时可直接「从备份恢复」。

Tablet & keyboard / 平板与键盘

• Two-pane list/detail layout on large screens, with a draggable divider and a collapsible session list. 大屏双栏布局，分隔线可拖动，会话列表可收起。
• Hardware keyboard: shortcuts for New Chat and Search, Ctrl/arrow/F-keys in the terminal; mouse drag-to-select with a right-click menu. 硬件键盘：新建会话与搜索快捷键，终端支持 Ctrl、方向键和 F1–F12；鼠标拖选文字并支持右键菜单。

Floating capsule & notifications / 悬浮胶囊与通知

• A redesigned floating capsule showing the current tool, a streaming state, the reply excerpt when done, your Soul identity and a count when several tasks run. 悬浮胶囊重新设计：显示当前工具、输出中状态、完成后的回复摘要、Soul 身份，多任务时显示编号。
• Live Updates show the current phase: tool, thinking or compacting. 实时活动通知显示当前阶段：工具、思考或压缩中。

Sandbox & tools / 沙盒与工具

• Each command runs in its own process, so Stop really stops it. 每条命令在独立进程中运行，「停止」真正生效。
• Container Network DNS: follow the system or set custom servers. 容器网络 DNS：跟随系统或自定义服务器。
• tsnet-based tools (e.g. Tailscale) can start inside the sandbox. 基于 tsnet 的工具（如 Tailscale）可在沙盒中运行。
• A queued command waits for Android's per-app process limit to free up instead of being killed, and its result notes the wait. 命令会排队等待 Android 的单应用进程上限空出，不会被系统杀掉，结果中会注明排过队。

Other / 其他

• Custom Soul icon (emoji or image). 自定义 Soul 头像（Emoji 或图片）。
• APK preview shows package info; web preview shows real JS alert/confirm/prompt dialogs. APK 预览显示应用包信息；网页预览支持原生 JS 弹窗（alert / confirm / prompt）。
• Settings › Storage: multi-select Clear Files / Delete, log level setting and Clear Logs & Caches. 「设置 › 存储」支持多选清理或删除，可设置日志级别并一键清理日志与缓存。
• About shows the build date; user-installed CA certificates are supported for custom endpoints. 「关于」显示构建日期；自定义接口支持用户安装的 CA 证书。

Fixes / 修复

Stability & performance / 稳定性与性能

• Fixed a crash on every launch, crashes when a WebView renderer dies, a voice-input crash, and out-of-memory crashes with many sub agents. 修复每次启动即崩溃、WebView 渲染进程退出导致崩溃、语音输入崩溃，以及多个子代理时内存不足崩溃。
• Fixed stutter and heat from unbounded chat caching; chats are released under memory pressure. 修复会话缓存无上限导致的卡顿发热；内存紧张时自动释放会话。
• Streaming no longer rebuilds the whole message list per token, decorative animations share one throttled clock, and restore, Storage and logging are much faster. 流式输出不再每个字重建整个列表，装饰动画共用一个节流时钟；恢复、存储页和日志明显更快。
• Chats started at the same moment all start, keep their question on screen and get a title. 同时发起的多个会话都能正常开始，用户消息不再消失，标题正常生成。

Conversations / 会话

• A message sent while a task is running no longer vanishes; a background compaction no longer erases your draft. 任务运行中发送的消息不再消失；后台自动压缩不再清空输入框草稿。
• Edit, retry and delete-from-here act on the right message, so an edit no longer leaves the old text in the model's context. 编辑、重试、从此处删除会作用于正确的消息，编辑后旧内容不再残留在模型上下文中。
• Switching model mid-task no longer wedges the session or keeps retrying the old model. 任务中切换模型不再卡死会话，也不会继续重试旧模型。
• Oversized requests recover instead of wedging the session; the group's context limit is enforced and context size is calibrated per model. 请求过大时自动恢复而不是卡住会话；遵守模型组的上下文上限，并按模型校准上下文大小。
• Compaction is bounded in time, falls back to another model, and its summary actually reaches the model; network failures no longer force a compaction. 压缩有时间上限、可回退到其他模型，摘要真正传给模型；网络错误不再触发压缩。
• Streaming text no longer jitters, flickers, fades in twice, or stays invisible after a tool call. 流式文字不再抖动、闪烁、重复淡入，工具调用后也不会不显示。
• Opening a chat no longer pops up the keyboard; session groups start collapsed and open one at a time. 打开会话不再自动弹出键盘；会话分组冷启动时全部折叠，一次只展开一个。

Providers & models / 服务商与模型

• Fixed request errors with Gemini tools, DeepSeek HEIC photos, OpenAI Responses tool results, OpenRouter reasoning, Cerebras and Claude 5 thinking. 修复 Gemini 工具调用、DeepSeek 的 HEIC 图片、OpenAI Responses 工具结果、OpenRouter 推理、Cerebras 及 Claude 5 思考参数导致的请求错误。
• Images from tools reach the model, attachments survive re-editing, and non-vision models get the image's file path instead of a failure. 工具返回的图片能传给模型，重新编辑时保留附件，不支持图片的模型会收到图片路径而不是报错。
• OAuth credentials are never deleted on a rejected refresh; the provider is marked instead. OAuth 刷新被拒时不再删除凭据，改为标记该服务商。
• Keyless self-hosted providers can be refreshed and selected; a model skipped in one refresh keeps its overrides. 无需 API Key 的自建服务可正常刷新和选用；某次刷新中缺失的模型不会丢失自定义设置。
• A failed title model hands over to the next one instead of leaving "New Chat". 标题模型失败时自动换下一个模型，不再停在「New Chat」。

Sub agents / 子代理

• Custom sub agents no longer vanish on relaunch; stopping one stops its whole batch; a correction sent while it writes its final answer is no longer lost. 自定义子代理重启后不再消失；停止一个会停止整批；在其撰写最终回答时发出的纠正不再丢失。

Backup / 备份

• S3 signing, anonymous destinations, self-signed certificates and slow WebDAV servers work; environment variable values, thinking levels, sub agents and compaction markers survive a restore. 修复 S3 签名、匿名目标、自签名证书与慢速 WebDAV；恢复后环境变量值、思考强度、子代理和压缩标记都能保留。

Browser, tools & sandbox / 浏览器、工具与沙盒

• Each chat drives its own browser tabs; password fields are redacted from page snapshots; frozen tabs recover; screenshots include WebGL content. 每个会话使用自己的浏览器标签页；页面快照中的密码字段会被遮盖；卡死的标签页自动恢复；截图包含 WebGL 内容。
• A command that starts an MCP daemon returns as soon as it finishes; timed-out commands are killed with their whole process tree. 启动 MCP 守护进程的命令执行完即返回；超时命令会连同整个进程树一起结束。
• Storage no longer reports inflated sizes (e.g. 43.89 GB for the shell container). 存储空间不再虚高（如 Shell 容器显示 43.89 GB）。
• Scheduled task alarms are re-registered after an app update or force-stop. 应用更新或强制停止后自动重新注册定时任务闹钟。
• Calendar writes target the right calendar and store all-day events on the right day. 日历写入指定的日历，全天事件存到正确日期。

Security & privacy / 安全与隐私

• OAuth codes and verifiers are no longer logged; backup passphrase fields no longer teach the keyboard. 不再记录 OAuth 授权码与校验值；备份密码输入框不再被输入法学习。

Polish / 视觉打磨

• Consistent 48 dp buttons and equal-height settings rows across the app. 全局统一 48dp 按钮高度，设置项行高一致。
• The chat "…" menu uses consistent outlined icons; the app follows your rotation setting. 聊天「…」菜单图标统一为线框风格；屏幕方向跟随系统设置。
• Markdown: aligned table grid lines, even inline-code backgrounds, correct line spacing and equation numbers. Markdown：表格线对齐、行内代码背景高度一致、行距与公式编号显示正确。

Full changelog: https://github.com/OpenMinis/OpenMinis/compare/1.13...1.14