💖🧸 自托管、归你拥有的 Grok 风格 AI 伴侣与 waifu / 赛博生命灵魂容器，目标是接近 Neuro-sama 的高度；支持实时语音聊天、Minecraft 和 Factorio 游玩，支持 Web / macOS / Windows。

文件	最后提交记录	最后更新时间
.agents	chore(.agents/skills): updated agent-browser-electron	1 个月前
.cursor	feat(ci): lint swift code	6 个月前
.gemini	chore: disable gemini review	1 个月前
.github	test: resolve spine typecheck and unit test failures in CI (#1895)	1 个月前
.vscode	chore(ci): added Godot vscode extension recommendation, and .prototools for managing dotnet versions	2 个月前
.zed	chore(workspace): added Zed editor config	5 个月前
apps	test: resolve spine typecheck and unit test failures in CI (#1895)	1 个月前
bucket	fix(bucket): update Scoop manifest to v0.9.0-alpha.18 (#1430) --------- Co-authored-by: Matt Van Horn <455140+mvanhorn@users.noreply.github.com> Co-authored-by-agent: Claude Opus 4.6 <noreply@anthropic.com>	3 个月前
docs	🖼️ docs(readme): use PNG download buttons on GitCode	1 个月前
engines	fix(stage-ui-three): OrbitControl initialisation (#1877) ## Summary - Fix the first Live2D -> VRM switch camera regression where OrbitControls could clamp the camera into the model. - Remove the VRM render scale control from the model settings panel. - Remove the now-unused `settings.vrm.render-scale` locale entries. - Ignore Godot/.NET `bin/` and `obj/` build artifacts under `engines/stage-tamagotchi-godot`. ## Root Cause On the first VRM load, `OrbitControls` can mount before the VRM bootstrap writes a real `modelSize`. The previous watcher copied `modelSize.z = 0` into `minDistance` and `maxDistance`, so later camera updates were clamped onto the target before the bootstrap camera pose finished applying. ## Changes - Skip OrbitControls distance-bound updates until `modelSize.z` is finite and non-zero. - Keep the render scale runtime state intact, but remove the VRM settings UI entry so it is no longer hot-updated from the panel. - Add Godot project ignores for local .NET build output. --------- Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>	1 个月前
integrations	release: v0.10.2	1 个月前
nix	chore(ci): add stage-tamagotchi-godot in publish (#1841)	1 个月前
packages	test: resolve spine typecheck and unit test failures in CI (#1895)	1 个月前
patches	chore(stage-tamagotchi): patch uiohook-napi to use prebuilt binaries and skip postinstall build	1 个月前
plugins	release: v0.10.2	1 个月前
scripts	chore(ci): resolve OpenCollective 401 in sponsors workflow (#1843)	1 个月前
services	fix(computer-use-mcp): bound waitForElement frame timeouts (#1856) ## Summary\n- Pass the remaining waitForElement budget into each frame-level CU_ACTION send so unresponsive frames cannot consume the fixed 8s sendMessage timeout.\n- Use the remaining deadline for each poll and stop polling immediately when the budget is exhausted.\n- Reduce the bridge-side waitForElement grace from the legacy 9.5s buffer to a small transport grace.\n- Add a regression test covering the hanging-extension case.\n\n## Validation\n- pnpm -C services/computer-use-mcp exec vitest run --config ./vitest.config.ts src/browser-dom/extension-bridge.test.ts --------- Co-authored-by: Neko <neko@ayaka.moe> Co-authored-by: 刘梓恒 <160735726+3361559784@users.noreply.github.com> Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>	1 个月前
.dockerignore	feat(ci): proper Dockerfile for current project setup & release workflow Close #424	10 个月前
.editorconfig	feat(stage-tamagotchi): add click through support (previous macOS only, now both supported) (#187) --------- Co-authored-by: DWCarrot <23579844+DWCarrot@users.noreply.github.com> Co-authored-by: Makito <5277268+sumimakito@users.noreply.github.com> Co-authored-by: Midori Kochiya <1381736+kmod-midori@users.noreply.github.com>	1 年前
.gitattributes	fix(ci): min-integrity: none for safeoutput mcp call (#1645)	2 个月前
.gitignore	feat(stage-ui): chat sync (#1775)	1 个月前
.prototools	chore(ci): added Godot vscode extension recommendation, and .prototools for managing dotnet versions	2 个月前
.tool-versions	chore(workspace): bump node.js version to 24.13.0	3 个月前
AGENTS.md	refactor(stage-web,stage-ui): replace `warpdrive` plugin with `unplugin-basemove` (#1818)	1 个月前
CLAUDE.md	chore(docs): replace CLAUDE.md symlink with @AGENTS.md reference (#1455)	3 个月前
LICENSE	docs: update LICENSE	1 年前
README.md	📝 docs(readme): center GitCode banner image	1 个月前
bump.config.ts	chore(bumpp): bump versions in iOS project	3 个月前
crowdin.yml	chore(ci): autofix crowdin files [skip ci]	7 个月前
cspell.config.yaml	feat(stage-tamagotchi,stage-shared): listen keyup for global shortcut with uiohook	1 个月前
default.nix	feat(nix): add nix flake and package (#319) * feat(nix): add nix flake and package * feat: support darwin --------- Co-authored-by: Neko <neko@ayaka.moe>	11 个月前
eslint.config.js	feat(auth): email login & profile (#1745) Co-authored-by: Liet Blue <127093491+lietblue@users.noreply.github.com>	2 个月前
flake.lock	chore(ci): Remove CI concurrency group & small fixes (#1758) ## Description <!-- Please insert your description here and especially provide info about the "what" this PR is solving --> It has no effect. The actions are run in parallel anyway. They have been previously updated to not fail when run in parallel. ## Linked Issues <!-- Optional, if you have any --> ## Additional Context <!-- e.g. is there anything you'd like reviewers to focus on? -->	2 个月前
flake.nix	chore(ci): Remove CI concurrency group & small fixes (#1758) ## Description <!-- Please insert your description here and especially provide info about the "what" this PR is solving --> It has no effect. The actions are run in parallel anyway. They have been previously updated to not fail when run in parallel. ## Linked Issues <!-- Optional, if you have any --> ## Additional Context <!-- e.g. is there anything you'd like reviewers to focus on? -->	2 个月前
knip.json	perf: setup knip, remove unused dependencies (#992)	5 个月前
package.json	test: resolve spine typecheck and unit test failures in CI (#1895)	1 个月前
pnpm-lock.yaml	test: resolve spine typecheck and unit test failures in CI (#1895)	1 个月前
pnpm-workspace.yaml	chore(stage-tamagotchi): patch uiohook-napi to use prebuilt binaries and skip postinstall build	1 个月前
posthog.config.ts	feat(inference): unify and optimize WebGPU inference pipeline (#1622) ## Problem The WebGPU inference pipeline had several structural issues: 1. No unified protocol — Kokoro TTS, Whisper ASR, and Background Removal workers each used their own ad-hoc message formats. Adding a new model meant reinventing worker communication from scratch. 2. Infrastructure existed but was disconnected — `GPUResourceCoordinator`, `LoadQueue`, `InferenceWorkerManager`, and `protocol.ts` were all implemented but had zero consumers. The adapters duplicated the same lifecycle/timeout/mutex patterns independently. 3. Performance gaps — Kokoro only offered fp32 on WebGPU (no fp16), Whisper warm-up compiled shaders for 187.5s of dummy audio, audio transfer went through unnecessary WAV blob encode/decode, and `listVoices` reloaded the model every time. 4. Silent failures — Whisper worker's `generate()` had no try-catch; errors were swallowed and the main thread waited until timeout. 5. No graceful degradation — Whisper and Background Removal workers hardcoded `device: 'webgpu'` with no WASM fallback. 6. No observability — Only Kokoro had performance tracing. No adapter reported status to `useInferenceStatus`. No cache management UI existed. 7. Dead code accumulation — Old `KokoroWorkerManager` (232 lines), legacy Whisper message types, and scattered duplicate constants. ## Changes ### Phase 0 — Critical Performance & Bugs - Add `fp16-webgpu` dtype for Kokoro TTS (~2x inference speed on supported GPUs) - Fix Whisper warm-up tensor from `[1, 128, 3000]` → `[1, 128, 1]` (minimal shader compilation) - Fix Whisper worker silent error bug (add try-catch to `generate()` and `load()`) ### Phase 1 — Data Transfer & Caching - Switch Kokoro audio to Float32Array transferable (skip WAV blob encode in worker, lightweight WAV encode on main thread) - Cache `listVoices` results (skip redundant model reload when adapter state is `ready`) - Normalize progress reporting to 0-100 across all adapters, differentiate `warmup` phase ### Phase 2 — Protocol Unification & Infrastructure - Migrate all 3 workers + 3 adapters to unified `protocol.ts` message types (`load-model`, `run-inference`, `model-ready`, `inference-result`, `progress`, `error`) - Wire `GPUResourceCoordinator` into all adapters (VRAM allocation tracking, LRU ordering, memory pressure events) - Wire `LoadQueue` into all adapters (priority-based sequential model loading: TTS=10 > ASR=5 > BG_REMOVAL=1) - Add `coordinator.ts` global singleton for GPU coordinator + load queue - Add WebGPU detection + WASM fallback in Whisper and Background Removal workers ### Phase 3 — Error Recovery & Observability - Add restart logic with exponential backoff to Whisper adapter (matching Kokoro's existing pattern) - Integrate `classifyError()` (OOM / DEVICE_LOST / TIMEOUT classification) in Whisper adapter - Extend `defaultPerfTracer` to Whisper `transcribe()` and Background Removal `processImage()` - Wire `useInferenceStatus` into all 3 adapters (downloading → ready → terminated lifecycle) ### Phase 4 — Tests - Add unit tests for `AsyncMutex` (4 tests), `LoadQueue` (4 tests), `GPUResourceCoordinator` (7 tests) — all 15 passing ### Phase 5 — Cleanup & Features - Delete old `KokoroWorkerManager` (232 lines, zero consumers) - Delete orphaned `libs/workers/types.ts` (old Whisper message types) - Clean up `workers/kokoro/types.ts` (remove legacy message types, keep domain types) - Create centralized `constants.ts` (MODEL_IDS, MODEL_NAMES, TIMEOUTS, MAX_RESTARTS) - Remove hardcoded WebGPU check from background-removal devtools pages (worker auto-detects) - Add `useModelPreload` composable for generic idle-time preloading - Add `useInferencePreload` composable that reads provider config and preloads configured local models - Wire preloading into both `stage-web` and `stage-tamagotchi` App.vue (Kokoro TTS preloads 3s after init) - Add `ModelCacheManager.vue` settings component (cache size display, per-model status, clear cache) - Document GPU Device isolation architecture in protocol.ts ## After This PR - All inference workers speak the same protocol → adding a new model adapter is straightforward - GPU memory is tracked across all models with automatic pressure warnings at 80%/95% of VRAM budget - Models load sequentially via priority queue → no bandwidth/VRAM contention - Workers auto-detect WebGPU and fall back to WASM → works on browsers without WebGPU - Kokoro TTS preloads during idle time → "instant" first use for configured users - All adapters auto-restart on worker crashes (max 3 attempts, exponential backoff) - 15 unit tests cover core infrastructure (mutex, queue, coordinator) - Zero dead code remains in the inference pipeline ## Test Plan - [x] `pnpm exec vitest run packages/stage-ui/src/libs/inference/` — 15 tests pass - [x] `pnpm -F @proj-airi/stage-ui exec tsc --noEmit` — no TypeScript errors - [x] `pnpm lint:fix` — no lint errors in changed files - [ ] Manual: verify Kokoro TTS works with fp16-webgpu on a supported browser - [ ] Manual: verify Whisper ASR loads and transcribes correctly - [ ] Manual: verify Background Removal works in devtools page - [ ] Manual: verify preloading triggers in console (`[Preload] Loading kokoro-...`) --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>	2 个月前
rustfmt.toml	chore: update rust fmt rules (#235)	1 年前
skills-lock.json	chore(.agent): added agent-browser skill	2 个月前
sponsorkit.config.js	feat(docs): sync download buttons and added SponsorKit (#1407)	3 个月前
tsconfig.json	chore(pocket): exclude build and public folder in tsconfig	2 个月前
turbo.json	feat(auth): OIDC (#1531)	3 个月前
uno.config.ts	feat(inference): unify and optimize WebGPU inference pipeline (#1622) ## Problem The WebGPU inference pipeline had several structural issues: 1. No unified protocol — Kokoro TTS, Whisper ASR, and Background Removal workers each used their own ad-hoc message formats. Adding a new model meant reinventing worker communication from scratch. 2. Infrastructure existed but was disconnected — `GPUResourceCoordinator`, `LoadQueue`, `InferenceWorkerManager`, and `protocol.ts` were all implemented but had zero consumers. The adapters duplicated the same lifecycle/timeout/mutex patterns independently. 3. Performance gaps — Kokoro only offered fp32 on WebGPU (no fp16), Whisper warm-up compiled shaders for 187.5s of dummy audio, audio transfer went through unnecessary WAV blob encode/decode, and `listVoices` reloaded the model every time. 4. Silent failures — Whisper worker's `generate()` had no try-catch; errors were swallowed and the main thread waited until timeout. 5. No graceful degradation — Whisper and Background Removal workers hardcoded `device: 'webgpu'` with no WASM fallback. 6. No observability — Only Kokoro had performance tracing. No adapter reported status to `useInferenceStatus`. No cache management UI existed. 7. Dead code accumulation — Old `KokoroWorkerManager` (232 lines), legacy Whisper message types, and scattered duplicate constants. ## Changes ### Phase 0 — Critical Performance & Bugs - Add `fp16-webgpu` dtype for Kokoro TTS (~2x inference speed on supported GPUs) - Fix Whisper warm-up tensor from `[1, 128, 3000]` → `[1, 128, 1]` (minimal shader compilation) - Fix Whisper worker silent error bug (add try-catch to `generate()` and `load()`) ### Phase 1 — Data Transfer & Caching - Switch Kokoro audio to Float32Array transferable (skip WAV blob encode in worker, lightweight WAV encode on main thread) - Cache `listVoices` results (skip redundant model reload when adapter state is `ready`) - Normalize progress reporting to 0-100 across all adapters, differentiate `warmup` phase ### Phase 2 — Protocol Unification & Infrastructure - Migrate all 3 workers + 3 adapters to unified `protocol.ts` message types (`load-model`, `run-inference`, `model-ready`, `inference-result`, `progress`, `error`) - Wire `GPUResourceCoordinator` into all adapters (VRAM allocation tracking, LRU ordering, memory pressure events) - Wire `LoadQueue` into all adapters (priority-based sequential model loading: TTS=10 > ASR=5 > BG_REMOVAL=1) - Add `coordinator.ts` global singleton for GPU coordinator + load queue - Add WebGPU detection + WASM fallback in Whisper and Background Removal workers ### Phase 3 — Error Recovery & Observability - Add restart logic with exponential backoff to Whisper adapter (matching Kokoro's existing pattern) - Integrate `classifyError()` (OOM / DEVICE_LOST / TIMEOUT classification) in Whisper adapter - Extend `defaultPerfTracer` to Whisper `transcribe()` and Background Removal `processImage()` - Wire `useInferenceStatus` into all 3 adapters (downloading → ready → terminated lifecycle) ### Phase 4 — Tests - Add unit tests for `AsyncMutex` (4 tests), `LoadQueue` (4 tests), `GPUResourceCoordinator` (7 tests) — all 15 passing ### Phase 5 — Cleanup & Features - Delete old `KokoroWorkerManager` (232 lines, zero consumers) - Delete orphaned `libs/workers/types.ts` (old Whisper message types) - Clean up `workers/kokoro/types.ts` (remove legacy message types, keep domain types) - Create centralized `constants.ts` (MODEL_IDS, MODEL_NAMES, TIMEOUTS, MAX_RESTARTS) - Remove hardcoded WebGPU check from background-removal devtools pages (worker auto-detects) - Add `useModelPreload` composable for generic idle-time preloading - Add `useInferencePreload` composable that reads provider config and preloads configured local models - Wire preloading into both `stage-web` and `stage-tamagotchi` App.vue (Kokoro TTS preloads 3s after init) - Add `ModelCacheManager.vue` settings component (cache size display, per-model status, clear cache) - Document GPU Device isolation architecture in protocol.ts ## After This PR - All inference workers speak the same protocol → adding a new model adapter is straightforward - GPU memory is tracked across all models with automatic pressure warnings at 80%/95% of VRAM budget - Models load sequentially via priority queue → no bandwidth/VRAM contention - Workers auto-detect WebGPU and fall back to WASM → works on browsers without WebGPU - Kokoro TTS preloads during idle time → "instant" first use for configured users - All adapters auto-restart on worker crashes (max 3 attempts, exponential backoff) - 15 unit tests cover core infrastructure (mutex, queue, coordinator) - Zero dead code remains in the inference pipeline ## Test Plan - [x] `pnpm exec vitest run packages/stage-ui/src/libs/inference/` — 15 tests pass - [x] `pnpm -F @proj-airi/stage-ui exec tsc --noEmit` — no TypeScript errors - [x] `pnpm lint:fix` — no lint errors in changed files - [ ] Manual: verify Kokoro TTS works with fp16-webgpu on a supported browser - [ ] Manual: verify Whisper ASR loads and transcribes correctly - [ ] Manual: verify Background Removal works in devtools page - [ ] Manual: verify preloading triggers in console (`[Preload] Loading kokoro-...`) --------- Co-authored-by: Claude Opus 4.6 (1M context) <noreply@anthropic.com> Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com>	2 个月前
vite-env.d.ts	refactor(stage-*): analytics and telemetry, including consent (#1414) --------- Co-authored-by-agent: Codex <267193182+codex@users.noreply.github.com> Co-authored-by-agent: gemini-code-assist[bot] <176961590+gemini-code-assist[bot]@users.noreply.github.com> Co-authored-by: autofix-ci[bot] <114827586+autofix-ci[bot]@users.noreply.github.com> Co-authored-by: RainbowBird <rbxin2003@outlook.com> Co-authored-by: Neko <neko@ayaka.moe>	3 个月前
vitest.config.ts	refactor(stage-web,stage-ui): replace `warpdrive` plugin with `unplugin-basemove` (#1818)	1 个月前

Project AIRI

复刻 Neuro-sama，让 AI waifu / 虚拟角色也能来到我们的世界。

[加入 Discord] [试试看] [English] [日本語] [Русский] [Tiếng Việt] [Français] [한국어]

深受 Neuro-sama 启发

Warning

注意： 我们没有发行任何与本项目关联的加密货币或代币，请注意判断资讯并谨慎行事。

Note

我们有一个专门的组织 @proj-airi 用于所有从 Project AIRI 诞生的子项目，快来看看吧！

RAG（检索增强生成）、记忆系统、嵌入式数据库、图标、Live2D 实用工具等等！

Tip

我们在 Crowdin 上有翻译项目。如果你发现翻译不准确，欢迎在上面参与翻译并贡献修正。

你是否梦想过拥有一个赛博生命（赛博 waifu、数字桌宠），或者能与你玩耍和交谈的数字伴侣？

借助现代大型语言模型的力量，像是 ChatGPT 和著名的 Claude 所能带来的，想要 LLM（大语言模型）和我们角色扮演、聊天已经超简单了，每个人都能上手。而像 Character.ai（又称 c.ai）和 JanitorAI 这样的平台，以及本地应用如 SillyTavern（又称酒馆），已经是基于聊天或文字冒险游戏体验的相当不错的解决方案。

但是，如何赋予它们玩游戏的能力呢？让它们能看到你正在编写的代码？不仅能一边聊天一边玩游戏，也可以看视频，还能做很多其他事情？

你可能已经知道 Neuro-sama，她目前是最好的能够玩游戏、聊天并与你和参与者（在VTuber社区中）互动的 AI VTuber / 伴侣，有些人也称这种存在为"数字人"。可惜的是，她并不开源，当她从直播中下线后，你就无法与她互动了。

因此，这个项目 AIRI，在这里提供了另一种可能性：让你轻松拥有自己的数字生命、赛博生命，随时随地。

我们发布的 DevLog & 近期更新

DevLog @ 2026.01.01 — 2026 年 1 月 1 日
DevLog @ 2025.10.20 — 2025 年 10 月 20 日
DevLog @ 2025.08.05 — 2025 年 8 月 5 日
DevLog @ 2025.08.01 — 2025 年 8 月 1 日
DevLog @ 2025.07.18 — 2025 年 7 月 18 日
DreamLog 0x1 — 2025 年 6 月 16 日
...更多请见文档站

这个项目有什么特别？

与其他 AI 和 LLM 驱动的 VTuber 开源项目不同，アイリ VTuber 从开始开发的第一天开始就支持多种 Web 技术，涵盖诸如 WebGPU、WebAudio、Web Workers、WebAssembly、WebSocket 等已经广泛应用或仍在大量实验的 API。

Tip

担心使用 Web 技术会导致性能下降吗？

不用担心，浏览器版本旨在展示我们在浏览器和 webview 里能做到什么，但不会完全依赖它。AIRI 桌面版默认原生支持 NVIDIA CUDA 和 Apple Metal（多亏了 HuggingFace 与可爱的 candle 项目），且无需复杂的依赖管理。考虑到取舍，我们在图形、布局、动画和 WIP 插件系统上部分使用了 Web 技术，方便大家集成。

这意味着 アイリ VTuber 能够在现代浏览器和设备上运行，甚至能够在移动设备上运行（已经完成了 PWA 支持），这为我们（贡献者们）带来了更多的可能性，让我们得以更进一步构建和扩展アイリ VTuber 的外部功能，而与此同时也不失配置的灵活性——可以有选择地在不同设备上启用需要 TCP 连接或其他非 Web 技术的功能，例如连接到 Discord 的语音频道一起开黑，或是和朋友们一起玩 Minecraft（我的世界）、Factorio（异星工厂）。

Note

アイリ VTuber 仍处于早期开发阶段，我们欢迎优秀的开发者加入我们，一起将它变为现实。

即使不熟悉 Vue.js、TypeScript 和所需的其他开发工具也没关系，我们也欢迎艺术家、设计师、运营策划的加入，你甚至可以成为第一个用アイリ VTuber 直播的博主。

如果你使用的是 React、Svelte，甚至 Solid 也没关系，您可以自己创建一个子目录，添加您希望在アイリ VTuber 中看到的功能，或者想实验的功能。

我们非常期待以下领域的朋友加入：

Live2D 模型师
VRM 模型师
VRChat 模型设计师
计算机视觉（CV）
强化学习（RL）
语音识别
语音合成
ONNX 推理运行时
Transformers.js
vLLM
WebGPU
Three.js
WebXR (也可以看看我们在 @moeru-ai 组织下另外的这个项目)

如果你已经感兴趣了，为什么不来这里和大家打个招呼呢？Would like to join part of us to build AIRI?

当前进度

目前已经能做到：

大脑
- 玩 Minecraft
- 玩 Factorio（进行中，但已提供 PoC 和 demo）
- 在 Telegram 聊天
- 在 Discord 聊天
- 记忆
  - 纯浏览器内数据库支持（DuckDB WASM | pglite）
  - Alaya 记忆层（施工中）
- 纯浏览器的本地推理（基于 WebGPU）
耳朵
- 浏览器音频输入
- Discord 音频输入
- 客户端语音识别
- 客户端说话检测
嘴巴
- ElevenLabs 语音合成
身体
- VRM 支持
  - 控制 VRM 模型
- VRM 模型动画
  - 自动眨眼
  - 自动看
  - 空闲眼睛移动
- Live2D 支持
  - 控制 Live2D 模型
- Live2D 模型动画
  - 自动眨眼
  - 自动看
  - 空闲眼睛移动

开发

开发本项目的详细指南请参阅 CONTRIBUTING.md

Note

默认情况下 pnpm dev 会启动 Stage Web（浏览器版）的开发服务器；如果你想尝试桌面版，请先阅读 CONTRIBUTING.md 正确完成环境配置。

pnpm i
pnpm dev

网页版 (也就是 airi.moeru.ai 的版本)

pnpm dev

桌面版（也叫拓麻歌子，aka 电子宠物）

pnpm dev:tamagotchi

我们提供了拓麻歌子的 Nix 包。先启用 flakes，然后可以直接运行：

nix run github:moeru-ai/airi

文档站

pnpm dev:docs

发布

运行 bumpp 来更新 monorepo 版本：

npx bumpp --no-commit --no-tag

原生支持的 LLM API 服务来源列表（由 xsai 驱动）

从这个项目诞生的子项目

Awesome AI VTuber：AI VTuber 及相关项目的精选列表
unspeech：通用的 /audio/transcriptions 与 /audio/speech 代理，类似 LiteLLM，但适用于任意 ASR 和 TTS
hfup：帮助部署与打包到 HuggingFace Spaces 的工具
xsai-transformers：面向 xsAI 的实验性 🤗 Transformers.js provider
WebAI: Realtime Voice Chat：从零实现 ChatGPT 实时语音（VAD + STT + LLM + TTS）的完整示例
@proj-airi/drizzle-duckdb-wasm：DuckDB WASM 的 Drizzle ORM 驱动
@proj-airi/duckdb-wasm：好用的 @duckdb/duckdb-wasm 封装
AIRI Factorio: 让 AIRI 玩 Factorio
Factorio RCON API: Factorio 无头服务器控制台的 RESTful API 封装
autorio: Factorio 自动化库
tstl-plugin-reload-factorio-mod: 开发时支持热重载 Factorio 模组
Velin：用 Vue SFC 和 Markdown 写可管理的多轮提示词
demodel：轻松加速不同推理引擎/下载器拉取模型与数据集
inventory：集中式模型目录与默认 provider 配置的后端服务
MCP Launcher：易用的 MCP 构建器与启动器，就像模型界的 Ollama！
🥺 SAD：自托管与浏览器运行 LLM 的文档与笔记

%%{ init: { 'flowchart': { 'curve': 'catmullRom' } } }%%

flowchart TD
  Core("Core")
  Unspeech("unspeech")
  DBDriver("@proj-airi/drizzle-duckdb-wasm")
  MemoryDriver("[WIP] Memory Alaya")
  DB1("@proj-airi/duckdb-wasm")
  SVRT("@proj-airi/server-runtime")
  Memory("Memory")
  STT("STT")
  Stage("Stage")
  StageUI("@proj-airi/stage-ui")
  UI("@proj-airi/ui")

  subgraph AIRI
    DB1 --> DBDriver --> MemoryDriver --> Memory --> Core
    UI --> StageUI --> Stage --> Core
    Core --> STT
    Core --> SVRT
  end

  subgraph UI_Components
    UI --> StageUI
    UITransitions("@proj-airi/ui-transitions") --> StageUI
    UILoadingScreens("@proj-airi/ui-loading-screens") --> StageUI
    FontCJK("@proj-airi/font-cjkfonts-allseto") --> StageUI
    FontXiaolai("@proj-airi/font-xiaolai") --> StageUI
  end

  subgraph Apps
    Stage --> StageWeb("@proj-airi/stage-web")
    Stage --> StageTamagotchi("@proj-airi/stage-tamagotchi")
    Core --> RealtimeAudio("@proj-airi/realtime-audio")
    Core --> PromptEngineering("@proj-airi/playground-prompt-engineering")
  end

  subgraph Server_Components
    Core --> ServerSDK("@proj-airi/server-sdk")
    ServerShared("@proj-airi/server-shared") --> SVRT
    ServerShared --> ServerSDK
  end

  STT -->|Speaking| Unspeech
  SVRT -->|Playing Factorio| F_AGENT
  SVRT -->|Playing Minecraft| MC_AGENT

  subgraph Factorio_Agent
    F_AGENT("Factorio Agent")
    F_API("Factorio RCON API")
    factorio-server("factorio-server")
    F_MOD1("autorio")

    F_AGENT --> F_API -.-> factorio-server
    F_MOD1 -.-> factorio-server
  end

  subgraph Minecraft_Agent
    MC_AGENT("Minecraft Agent")
    Mineflayer("Mineflayer")
    minecraft-server("minecraft-server")

    MC_AGENT --> Mineflayer -.-> minecraft-server
  end

  XSAI("xsAI") --> Core
  XSAI --> F_AGENT
  XSAI --> MC_AGENT

  Memory_PGVector("@proj-airi/memory-pgvector") --> Memory

  style Core fill:#f9d4d4,stroke:#333,stroke-width:1px
  style AIRI fill:#fcf7f7,stroke:#333,stroke-width:1px
  style UI fill:#d4f9d4,stroke:#333,stroke-width:1px
  style Stage fill:#d4f9d4,stroke:#333,stroke-width:1px
  style UI_Components fill:#d4f9d4,stroke:#333,stroke-width:1px
  style Server_Components fill:#d4e6f9,stroke:#333,stroke-width:1px
  style Apps fill:#d4d4f9,stroke:#333,stroke-width:1px
  style Factorio_Agent fill:#f9d4f2,stroke:#333,stroke-width:1px
  style Minecraft_Agent fill:#f9d4f2,stroke:#333,stroke-width:1px

  style DBDriver fill:#f9f9d4,stroke:#333,stroke-width:1px
  style MemoryDriver fill:#f9f9d4,stroke:#333,stroke-width:1px
  style DB1 fill:#f9f9d4,stroke:#333,stroke-width:1px
  style Memory fill:#f9f9d4,stroke:#333,stroke-width:1px
  style Memory_PGVector fill:#f9f9d4,stroke:#333,stroke-width:1px