Skip to content

CanvasAnvil is an open source AI diagram maker and flowchart generator - an infinite-canvas diagram editor and self-hosted drawio alternative that writes editable .drawio files. Text to diagram, then fix nodes, layout and labels in place. Also AI presentations, interior design CAD/BOM, posters and infographics.

docs: drop the README hero image, release notes and WeChat group QR

codingFeng10127f1aea1b8a7
diff --git a/CHANGELOG.md b/CHANGELOG.md
deleted file mode 100644
index 955f50a..0000000
--- a/CHANGELOG.md
+++ /dev/null
@@ -1,218 +0,0 @@
-# Changelog
-
-All notable changes to this project will be documented in this file.
-
-The format is based on Keep a Changelog, and the project aims to follow Semantic Versioning.
-
-## [Unreleased]
-
-### Changed
-
-- Relicensed the project from AGPL-3.0 to MIT.
-
-## [2.2.0] - 2026-10-26
-
-A structural refactor of the whole project, a fourth canvas, and a pass over
-the whole interface: empty states, unified naming and styling, complete
-Chinese/English support, and a portal that lays out at every common window
-size.
-
-The refactor itself: the repository now separates
-`client/`, `server/`, `contracts/` and `resources/`, every workspace has the
-same shape, and the shared UI kit, storage layer, file pipeline and i18n each
-exist once instead of per workspace.
-
-### Removed
-
-- Removed the `pptist-lab` editor integration. The PPT canvas stays; editable decks export directly as `.pptx`.
-- Removed the poster, product and infographic canvases, narrowing the product to Flow, Interior Design and PPT. (Free Canvas was added afterwards; see Added.)
-- Removed the multi-provider model layer and the vendor picker. All model access goes through one OpenAI-compatible transport.
-- Removed the Next.js port that shadowed the Vite client, and roughly 900 lines of unreachable code the old `PptWorkspace` was hiding.
-- Removed the editable-export review and the pipeline behind it. Asking for an editable `.pptx` used to open a text-box confirmation pass and then spend three model calls per slide; it now writes the file from what the deck already has, in about a second. A deck that went through the old review still exports with its text editable, but new decks export as images -- which makes the editable and image exports the same thing.
-- Removed the nine-vendor image registry the `cad-skill` and `ppt-skill` packages still carried, along with the `provider` config key it read. Eight of those vendors had no base URL to reach and each skill shipped its own byte-identical copy of the table. The endpoint now follows from the model's name -- `gpt-image-*` and `dall-e-*` use the images endpoint, everything else answers through chat -- and `baseUrl` points at whichever OpenAI-compatible gateway serves it.
-- Removed the settings button and dialog from the Diagram chat panel; the top-bar settings cover the model configuration. The access-code field, the draw.io sketch style and the leave-page confirmation went with it.
-- Hid the "Export editable PPTX" menu entry. The export code, including the on-device OCR rebuild, is kept behind a flag in `DeckView.tsx`.
-- Removed unused code and assets: two UI components, three dependencies (`@radix-ui/react-label`, `@radix-ui/react-switch`, `vite-tsconfig-paths`), about fifteen dead exports, 35 unused translation keys, the brush cursor, the svgedit builds, demo extension and source maps the CAD editor never loads (18 MB down to 4 MB), 30 example images nothing referenced (42 MB down to 12 MB), and the `/api/verify-access-code` and `/api/log-feedback` endpoints, which nothing called any more.
-
-### Added
-
-- Added **Free Canvas**, an Excalidraw board for images: generate a picture from a sentence, mark up what to change with arrows, circles or notes, and get a revised version beside the original.
-- Added a version strip to Free Canvas: thumbnails of every version of the selected picture, with what changed and when. New versions are placed to the right of the picture they came from.
-- Added empty states to all four chat panels: what the assistant does, and three example requests that fill the input when clicked.
-- Added a Chinese/English toggle to the workspace top bar, and a primary "Start creating" action to the portal homepage.
-- Added keyboard paging (arrow keys, PageUp/PageDown) to the portal, alongside the wheel.
-- Added a loading skeleton shaped like the workspace layout, in place of a lone spinner.
-- Added a **Present** button to the deck toolbar. The slideshow was complete but had no entry point, so full-screen presentation, arrow-key paging and Escape had all been unreachable.
-- Added a test suite (`npm test`, node:test) covering the pure logic the refactor had to preserve: AI request shaping, draw.io XML repair, chat storage, PPT persistence and editor adapters, the deck state machine, slide-version resolution, render-layer edits, material tokens and the chat message readers.
-
-### Changed
-
-- Renamed the Flow canvas to **Diagram** throughout: directories, identifiers, the workspace id, prompts and their `diagram_patch` payload type, and the `diagram-skill` bundle. A saved `"flow"` selection and `flow_patch` payloads in old chats are still read.
-- Unified the canvas names across the portal, the top bar and the chat panels (Diagram, Interior Design, PPT, Free Canvas); the CAD Assistant is now the Design Assistant.
-- PPT slide generation, slide editing and outline planning treat the template as a style and colour reference only; layouts follow each slide's content instead of repeating the template's.
-- Free Canvas annotation edits now render the marked-up picture at its native resolution, pass the annotation's written text verbatim, send the clean original alongside it, and ask for the exact old and new text of any text change.
-- Compressed the PPT templates from 285 MB of PNG to 34 MB of JPEG, with 480px thumbnails for the picker, loaded lazily.
-- Restyled the portal: one button style in place of the brush-stroke plate and brush cursor, drifting colour washes in place of stray decorative strokes, a two-row header below 1280px, and English copy that matches the Chinese.
-- Portal pages now scroll when they do not fit, and tighten their spacing on laptop-height windows so they fit on one screen at 1366x768 and 1280x800.
-- Disabled primary buttons are a neutral grey rather than half-transparent clay, and the PPT start button says why it is disabled.
-- The page's `lang` attribute now follows the interface language from the first paint.
-- `npm run lint` skips vendored bundles and allows the ideographic space in CJK regexes; it now reports no errors.
-- Turned on TypeScript strict mode and split the bundle; initial load dropped from 2.89 MB to roughly 700 KB raw.
-- Replaced the deck's 25 scattered step assignments with a state machine, so the creation step and its progress bar can no longer drift apart.
-- Consolidated the CAD and PPT chat panels, chat inputs and transcript message readers, which had been forks of one another.
-- Split the PPT canvas from one 5,300-line component into pure modules, per-feature hooks and one file per screen, and wrote that shape down in `docs/architecture.md` so it is a convention rather than one worked example.
-- Gave the CAD workspace one way to read SVG out of a model reply, and one ladder for applying a patch, replacing three drifted copies of the first and two identical copies of the second.
-
-### Fixed
-
-- Fixed the image sizes 256, 512 and 1792 being silently dropped by the server, which only accepted the gpt-image sizes.
-- Fixed Free Canvas messages showing raw base64 when the prompt contained a line break, `|` or `]`.
-- Fixed the Free Canvas annotation preview being lost on reload, and a reply that arrived after "clear" restoring the cleared chat and placing its picture on the empty board.
-- Fixed Free Canvas edits that failed while compositing the annotation leaving the user's message without a reply, and failure messages that hid the reason.
-- Fixed annotation notes written beside a picture disappearing from what the model saw, when the composite was re-encoded as JPEG with a transparent margin.
-- Fixed the Free Canvas version history going stale after a version was deleted, and its button being disabled with no explanation.
-- Fixed the portal's detail pages being cut off below 1280px wide, and in English at laptop heights, with no way to scroll to the hidden part.
-- Fixed the English homepage headline running "canvas" and "creation" together.
-- Fixed strings that stayed Chinese in English: the portal's preview captions, the Diagram upload checks, send/stop labels and error toasts, the PPT slide context menu and the "(empty)" marker in slide-change previews.
-- Fixed the chat input toolbar pushing the send button out of a narrow panel.
-- Fixed a stale test expecting uploaded templates without a category.
-- Fixed an infinite render loop in every workspace's chat panel. Two file-sync effects running one step apart traded two equal-but-distinct arrays back and forth on every render.
-- Fixed the theme and dark-mode switches in the Flow chat panel, which received their handlers as props and used none of them.
-- Fixed CAD patch failures being swallowed: an exception while applying a patch left no error and no retry, so the patch appeared to do nothing.
-- Fixed the SVG load-failure toast keeping the previous language after a language switch.
-- Fixed the export-review selection pointing at a text block from an image version no longer on screen.
-- Fixed attachment chips in the CAD and PPT transcripts showing an empty character count.
-- Fixed messages whose content is a multimodal part array reading as empty text in the CAD and PPT transcripts.
-- Fixed a Docker image that copied directories which no longer exist and omitted `resources/`, which the server reads at request time.
-- Fixed SVG arriving HTML-escaped being read as "no drawing here" by the CAD canvas, which silently dropped it. The three copies of that reader had drifted; only two decoded entities.
-- Fixed the CAD workspace preferring bare `<svg>` text over a declared patch payload, which returned markup with the JSON escaping still in it.
-- Fixed the recent-history context dropping the newest turns when over budget and keeping the oldest -- backwards for something the intent router reads for continuity.
-- Fixed nine slide edits in the outline editor mutating the state they were replacing, by copying the array but not the slide inside it.
-- Fixed `ppt-skill` accepting an unsupported image provider through configuration and failing later at the HTTP layer, where `cad-skill` rejects it up front.
-- Fixed text blocks read back from storage reaching the review canvas without their geometry, where they became boxes at undefined coordinates that could be neither seen nor selected.
-- Fixed the chat upload limits applying to one of the three shapes an attachment can arrive in, so a file sent as `image` or in a `content` array skipped both the size cap and the file count.
-- Fixed the Flow composer never growing past three lines. The shared `Textarea` did not forward its ref, so the height it measured on every keystroke was never applied and the box stayed at its minimum, with resizing disabled.
-- Fixed pictures attached in the chat panel reaching only the branch that edits an existing slide image. Every branch that redraws the slide dropped them -- which is the branch a request to add a picture takes, so the attachment was lost exactly when it was the point of the message. They now travel with all five generation and edit paths, named so the prompt can say where each one goes.
-- Fixed a slide's material images never reaching the image model. The picture was read from the description, collected and sent, then dropped by the server: the edits route attached only the first reference image, which the template had taken, and the generations route sends no images at all. Because the prompt still named each material and told the model to place it, the slide came back with an invented picture rather than a visible gap. Only the materials a description actually refers to are sent now, and image generation carries up to six references instead of sharing the vision cap of three.
-- Fixed the chosen template being outvoted by the outline. The planner had no parameter for a template and had never seen one, yet it wrote a background colour and art style into every slide's visual description; that concrete text then beat the template reference at render time, so a deck built on the dark navy template came out white and flat. The planner is now given the template image and writes descriptions that match its palette and art style, and the template still outranks a description that disagrees with it at render time.
-- Fixed a slide displaying with its text erased after an editable export, from the adapter preferring a render layer's background -- the text-stripped derivative -- over the one the deck view passed.
-
-### Security
-
-- Removed server-side debug logging that wrote entire conversations, including inline base64 images and everything the user typed, to the console on every request.
-- Prompt logging now writes only when `PROMPT_LOG_DIR` is set, instead of dropping a file containing the full prompt into the working directory by default.
-
-## [2.1.0] - 2026-04-25
-
-### Added
-
-- Added a standalone deployment path for `pptist-lab`, allowing the PPT editor to run as an independent service alongside the main CanvasAnvil app.
-- Added deployment documentation for the dual-service setup, covering the main application port and the dedicated PPT editor port.
-
-### Changed
-
-- Updated the CanvasAnvil portal interface with a refreshed entry experience and clearer product navigation.
-- Standardized the deployment ports around the main CanvasAnvil app and the standalone PPT editor service.
-- Updated and synchronized the online access URL to `https://canvasanvil.codingfgd.asia`.
-- Updated the PPT editor embedding flow so production deployments behind domain and HTTPS reverse proxy can resolve the correct parent window origin.
-
-### Fixed
-
-- Fixed production PPT editor embedding issues caused by mismatched `postMessage` origins when the main app is served from HTTPS default port and the PPT editor is served from a dedicated HTTPS port.
-
-## [2.0.0] - 2026-04-22
-
-### Added
-
-- Added three new canvases: `Poster`, `Infographic`, and `Product`.
-- Embedded `pptist-lab` directly inside the PPT workspace for editable PPT authoring.
-- Added a staged editable PPT review flow with automatic text-box detection, manual box correction, text extraction, and per-slide re-extraction.
-- Added local skill definitions for `Flow`, `CAD`, `PPT`, `Poster`, `Infographic`, and `Product` workflows under `skill/`.
-- Added configurable image-provider support for image-generation skills via `config/image-provider.json`, covering the supported built-in providers.
-- Added script-backed skill workflows for image generation, diagram export, CAD output bundling, PPT template generation, slide image rendering, PDF export, and image-based PPT export.
-
-### Changed
-
-- Promoted the release version from `1.0.2` to the next major release, `2.0.0`.
-- Expanded CanvasAnvil from `Flow | CAD | PPT` to `Flow | CAD | PPT | Poster | Infographic | Product`.
-- Updated editable PPT export to separate `text box review`, `text extraction`, and `start editing`.
-- Kept both the original slide image and the derived `textless background` version available during review.
-- Improved extracted text mapping so the embedded PPT editor keeps more of the original size, color, weight, spacing, alignment, and line-height.
-- Reduced repository documentation to English and Simplified Chinese only.
-- Standardized skill prompts and `SKILL.md` files around English instructions while keeping UI-language-specific output labels where needed.
-- Simplified the PPT skill export scope to `PDF` and image-based `PPT` deliverables.
-- Updated the CAD skill BOM output contract to produce `cad_bom.csv` instead of `cad_bom.json`.
-- Updated the Flow skill bundle workflow to skip standalone HTML preview generation by default.
-
-### Fixed
-
-- Fixed editable PPT export popup blocking by switching to an embedded editor flow.
-- Fixed oversized editable-export bootstrap payloads that previously overflowed browser storage.
-- Fixed review-mode state issues around text-box preparation, extraction progress, per-slide re-extraction visibility, and version switching.
-
-## [1.0.2] - 2026-04-15
-
-### Changed
-
-- Promoted the release version from `1.0.1` to `1.0.2`.
-- Switched PPT creation and in-editor iteration to an image-first workflow that regenerates slide images directly.
-- Split PPT export into `PDF`, image-based `PPT`, and `editable PPTX`, with OCR and text refill deferred to editable export only.
-- Refreshed the localized README files to reflect the new PPT workflow and current release version.
-
-### Fixed
-
-- Persisted uploaded PPT templates and hidden preset-template preferences through IndexedDB-backed local storage.
-
-## [1.0.1] - 2026-03-24
-
-### Changed
-
-- Promoted the release version from `1.0.0` to the patch release `1.0.1`.
-
-### Fixed
-
-- Moved oversized PPT workspace image persistence from `localStorage` to IndexedDB-backed storage to prevent refresh-time data loss.
-- Fixed PPT workspace restore order so lightweight snapshots no longer override the latest persisted image state.
-- Ensured PPT imported, edited, and generated slide images are converted to persistable URLs before being written into workspace state.
-- Moved CAD analysis and render image persistence to IndexedDB-backed storage with lightweight local snapshots for better refresh stability.
-- Prevented CAD render/analysis images from being dropped by temporary loading placeholders unless the user explicitly clears the workspace.
-
-## [1.0.0] - 2026-03-17
-
-### Added
-
-- Local PPT AI proxy route at `/api/ppt-ai` for browser-safe PPT chat, vision, and image workflows.
-- Major PPT release notes and release-version upgrade to `v1.0.0`.
-- Support for text editing workflows on PPT slides exported from NotebookLM.
-
-### Changed
-
-- Promoted the project from `0.1.2` to the first major usable release, `1.0.0`.
-- Refined PPT editing workflows to better support real-world slide refinement and text updates on existing decks.
-- PPT image uploads no longer inject raw base64 image payloads into chat text prompts.
-- PPT image generation and edit requests now use a local proxy path instead of direct browser-to-provider requests.
-- PPT image-reference handling now limits reference count, compresses large image inputs, and retries with a reduced reference set on failure.
-- PPT edit routing now treats uploaded images as optional materials unless the user explicitly asks to use them.
-- Improved PPT error handling for oversized requests, bad payloads, and provider/API failures.
-
-### Fixed
-
-- Resolved multiple garbled-text issues in the PPT workspace source.
-- Fixed local storage overflow issues caused by oversized PPT chat history payloads.
-- Reduced browser-side CORS failure exposure in PPT image workflows by routing through the local proxy.
-
-## [0.1.2] - 2026-03-06
-
-### Changed
-
-- CAD workflow now supports an analysis-first entry: requirement/planning -> analysis images -> 2D editing.
-- CAD 2D generation (`cad_svg_agent`) now consumes the two analysis images as visual references when available.
-- CAD analysis image agents now accept language as an explicit prompt variable and follow UI language settings.
-
-## [0.1.0] - 2026-02-19
-
-### Added
-
-- First open-source release.
-- Flow/CAD/PPT canvases with chat-driven generation and iterative edits.
-- Docker/Compose deployment support and Nginx reverse proxy configuration.
diff --git a/README.md b/README.md
index adaef6e..fbb1c57 100644
--- a/README.md
+++ b/README.md
@@ -1,7 +1,3 @@
-<p align="center">
- <img src="public/logo.png" alt="CanvasAnvil hero" width="92%" />
-</p>
-
<h1 align="center">CanvasAnvil</h1>
<p align="center">
@@ -131,35 +127,6 @@ A whiteboard for images: describe a picture in one sentence and it lands on the
| `PPT` | structured presentation generation and export | presentations, pitch decks, report slides |
| `Free Canvas` | text-to-image and edit-by-annotation on a whiteboard | posters, illustrations, revised versions of an image |
-## 🌈 Release
-
-Current release: `v2.2.0` (2026-10-26). See [CHANGELOG.md](CHANGELOG.md) for the full list.
-
-**Four canvases**
-
-- Diagram, Interior Design and PPT, plus the new **Free Canvas**: text-to-image on a whiteboard, editing by drawing on the picture, and a version strip for every image.
-- The Flow canvas is now called **Diagram** everywhere, from the UI to the code, prompts and the `diagram-skill` bundle. Saved selections and chat histories carry over.
-- The poster, product and infographic canvases and the embedded `pptist-lab` editor are gone. The editable-PPTX export is hidden from the menu for now; its code is kept.
-
-**A friendlier interface**
-
-- Every chat panel opens with a short guide and example requests instead of a blank column.
-- The homepage has a clear way in, one button style throughout, and lays out cleanly from 1024px laptop windows to large monitors, in both languages.
-- Switch between Chinese and English from the top bar of any workspace; every string, error message and screen-reader label follows.
-- Quieter details: neutral disabled buttons that say why they are disabled, loading skeletons, and gentle motion that respects the system's reduced-motion setting.
-
-**Better results**
-
-- Free Canvas reads the annotation at the picture's full resolution, takes written notes verbatim, and quotes the exact text to change — edits land where they were marked.
-- PPT slides take only the style and colours from the template, not its layout, so a deck no longer repeats one arrangement.
-- Image size options that the server used to drop silently now reach the model.
-
-**Lighter and cleaner**
-
-- PPT templates went from 285 MB to 34 MB, with small thumbnails for the picker; the bundled SVG editor from 18 MB to 4 MB; example images from 42 MB to 12 MB.
-- The repository separates `client/`, `server/`, `contracts/` and `resources/`, every workspace has the same three layers ([docs/architecture.md](docs/architecture.md)), TypeScript strict mode is on, and the initial load dropped from 2.89 MB to roughly 700 KB.
-- Over 450 tests (`npm test`) cover the pure logic; `npm run lint` reports no errors.
-
## 🌐 Online Access
Current online access URL:
@@ -255,9 +222,3 @@ See [LICENSE](LICENSE) for the full license text.
## 💌 Contact
-
-## 💬 Community
-
-欢迎加入「智构开源社」群聊,一起交流 CanvasAnvil 及开源项目相关内容。
-
-![智构开源社微信群二维码](docs/assets/wechat-group.png)
diff --git a/README.zh-CN.md b/README.zh-CN.md
index 33bf4c5..2e2b562 100644
--- a/README.zh-CN.md
+++ b/README.zh-CN.md
@@ -1,8 +1,4 @@
-<p align="center">
- <img src="public/logo.png" alt="CanvasAnvil 主视觉" width="92%" />
-</p>
-
-<h1 align="center">CanvasAnvil</h1>
+<h1 align="center">CanvasAnvil</h1>
<p align="center">
<strong>面向图表绘制、室内设计、PPT 演示与图片创作的 AI 多画布平台。</strong>
@@ -131,35 +127,6 @@ https://github.com/user-attachments/assets/582c70ae-38df-4949-913c-c3f4369a7c8c
| `PPT 演示` | 结构化演示文稿生成与导出 | 演示文稿、汇报幻灯片、提案稿 |
| `自由画布` | 白板上的文生图与按标注改图 | 海报、插画、同一张图的修改版本 |
-## 🌈 版本发布
-
-当前版本:`v2.2.0`(2026-10-26)。完整列表见 [CHANGELOG.md](CHANGELOG.md)。
-
-**四个画布**
-
-- 图表绘制、室内设计、PPT 演示之外,新增**自由画布**:白板上一句话生成图片,直接在图上画圈标注来改图,每张图都有版本条。
-- 原来的「流程画布」统一更名为**图表绘制(Diagram)**,从界面到代码、提示词和 `diagram-skill` 都已改名;已保存的选择和聊天记录会自动沿用。
-- 海报、产品、信息图画布和内嵌的 `pptist-lab` 编辑器已移除。「导出可编辑 PPTX」暂时从菜单里隐藏,相关代码保留。
-
-**更友好的界面**
-
-- 每个交互区打开时都有简短说明和可直接使用的示例,不再是一片空白。
-- 首页有了明确的入口,全站按钮风格统一;从 1024 宽的笔记本到大屏,中英文下都不再拥挤或被截断。
-- 任意工作区顶栏都能一键切换中英文,所有文字、报错和读屏标签都会跟着切换。
-- 细节打磨:禁用按钮改为中性灰并说明原因、加载时显示骨架屏、新增克制的动效,并遵守系统「减少动态效果」设置。
-
-**更准确的结果**
-
-- 自由画布按图片原始分辨率读取标注,标注里写的文字逐字传给模型,并要求引用原文和新文字,改图更准确。
-- PPT 生图只参考模板的风格与色调,不再照搬模板版式,整套 PPT 不会千篇一律。
-- 之前被服务端悄悄丢掉的图片尺寸选项,现在能正确传给模型。
-
-**更轻、更整洁**
-
-- PPT 模板图从 285 MB 压缩到 34 MB,模板列表改用缩略图;内置的 SVG 编辑器从 18 MB 精简到 4 MB;示例图从 42 MB 精简到 12 MB。
-- 仓库拆分为 `client/`、`server/`、`contracts/`、`resources/`,每个工作区结构一致(见 [docs/architecture.md](docs/architecture.md));开启 TypeScript 严格模式,首屏加载从 2.89 MB 降到约 700 KB。
-- 450 多个测试(`npm test`)覆盖核心逻辑,`npm run lint` 零错误。
-
## 🌐 在线使用
当前在线访问地址:
@@ -255,9 +222,3 @@ CanvasAnvil 采用 MIT 协议发布。
## 💌 联系方式
-
-## 💬 加入社区
-
-欢迎加入「智构开源社」群聊,一起交流 CanvasAnvil 及开源项目相关内容。
-
-![智构开源社微信群二维码](docs/assets/wechat-group.png)
diff --git a/docs/HANDOFF.md b/docs/HANDOFF.md
index c3a82ec..e2e08c7 100644
--- a/docs/HANDOFF.md
+++ b/docs/HANDOFF.md
@@ -1,6 +1,6 @@
# 交接文档
-> 版本 v2.2.0 · 2026-10-26。详细架构见 [architecture.md](architecture.md),完整变更见 [CHANGELOG.md](../CHANGELOG.md)。
+> 版本 v2.2.0 · 2026-10-26。详细架构见 [architecture.md](architecture.md)。
## 一句话
diff --git a/docs/assets/wechat-group.png b/docs/assets/wechat-group.png
deleted file mode 100644
index 2a33feb..0000000
Binary files a/docs/assets/wechat-group.png and /dev/null differ
diff --git a/public/logo.png b/public/logo.png
deleted file mode 100644
index 0390e91..0000000
Binary files a/public/logo.png and /dev/null differ