Compose

A lightweight alternative to Microsoft Word, with canvas design tools for reports, homework, posters and everyday documents.

Write like a document. Design like a canvas.

Windows preview: Download Compose · Windows guide. The portable ZIP is hosted on Cloudflare, includes a SHA-256 checksum, and needs no installer or account. The preview is unsigned; physical Windows graphics and interaction testing remains in progress.

Windows release refinements include platform-correct menu shortcuts, portable filenames, overwrite confirmation when adding .compose, a quiet GUI launch, and a clean ZIP with instructions and licenses. Undo/Redo and file/export commands now respect the unsaved-changes prompt; invalid inspector input stays editable after Tab. Native regression checks: cargo run -p compose -- --ui-smoke --release-checks --compact.

Compose combines rich text editing with movable images, shapes, layers and groups in a calm native interface. Built in Rust with GPUI and Parley, without Electron or webviews.

Compose is in active development. macOS is currently tested; Windows/Linux interaction checks are still pending; native Windows packaging is included. It uses its own editable .compose files in both modes. Word-compatible mode can export editable DOCX; DOCX import is not implemented yet.

Run

A native Android version with desktop icons, fonts and editable components is available too. Build an APK with ./scripts/package-android.sh; see Android instructions for supported devices, mobile features and emulator checks. It shares the Rust document engine and .compose files with desktop.

Install Rust (currently validated with 1.98.1). On macOS, install the Xcode Command Line Tools with xcode-select --install; the build uses Swift only for the small native printing adapter.

cargo run -p compose
# Open a saved document:
cargo run -p compose -- /path/to/document.compose

On macOS, create a local development app bundle:

./scripts/package-macos.sh
open dist/Compose.app

The bundle includes fonts and a vector app icon. It is locally ad-hoc signed, not a notarized distribution. macOS is the platform tested in this workspace. Linux/Windows support still needs platform-specific builds and interaction QA.

Working features

• Shape picker with Rectangle, Circle, Oval, Triangle, Diamond, Pentagon, Hexagon, Star, Arrow and Line. Draw or drag figures onto the page, or change a selected shape's type in the inspector. Circles stay proportionate; polygons support rounded corners, colors, borders and shadows.

• Two document modes: Word-compatible text/image layouts with editable DOCX export, and the full Compose canvas with shapes, frames, groups and effects.

• Six bundled typefaces with actual family previews, live search, real weights, italic variants where available, and seven typography presets.

• Rich text formatting applies only to the highlighted passage. Mixed sizes, family, weight, color, spacing, line height and alignment remain editable.

• Continuous HSV color picker with hex/RGBA and alpha input, 32 palette colors, and colors already used in the document. One color drag is one undo step.

• Assets library: Title, Callout, Pull quote, Key number, Feature card and Divider. Components insert as editable nested frames/text/shapes; they are independent layouts, not linked instances.

• Visual object fills, rounded corners, strokes and soft drop shadows, persisted in .compose files. Images retain original bytes and proportionate Scale mode.

• Layers tree with visibility, locks, nesting, renaming, ordering and duplication.

• Shift-click and marquee selection, shared movement/nudging, batch clipboard, grouping/ungrouping, alignment and even spacing for several objects.

• Multiple pages, page-size presets/orientation, editable document title, preview, pan/zoom, snapping, inspector fields and native menus.

• Undo/redo, native file dialogs, atomic saves and unsaved-change prompts.

• Export editable .compose copies and current-page PNGs at 1×, 2× or 300 DPI. Print all pages through the macOS print dialog, including its Save as PDF option.

• Copy/cut/paste layers and nested frames across documents in the same running app. Pasted images keep their original bytes and attribution.

• Copy highlighted passages with inline formatting; paste external text onto the canvas to create an editable text box.

• Replace images from your computer, web search or Commons while keeping their layer name, position, width and appearance. Scale adjusts height to the new proportions; Stretch keeps both dimensions.

Click the document title to rename it. The Assets tab holds editable layout blocks. The typeface field opens the font picker; the color row opens the palette and continuous picker. Use the eye/lock icons beside each layer. The preview triangle hides editing controls; Escape returns to the editor.

Double-click a blank area to write. T also lets you click and drag within an existing text box to highlight a passage. Double-click a word to select it. Bold, italic, underline, font size, family, and color apply only to that selection; the floating toolbar stays with the selected text. The inspector shows Mixed when selected characters have different property values.

Highlight a passage to open its centered text menu. Choose fonts, weights, exact sizes or presets, text styles, bold/italic/underline, color or clear formatting right beside the selection. Font search is focused automatically; Enter applies the first match. Styles include Title, Heading, Subheading, Body, Caption, Quote and Code. All changes apply to the highlighted passage. Run native interaction checks with cargo run -p compose -- --ui-smoke --text-menu.

Check every bundled native font face and the Light/Black controls with cargo run -p compose -- --ui-smoke --font-weights. The renderer maps typographic family names consistently between Parley and GPUI, including extended weights and italic variants.

Images default to Scale, which preserves their original proportions with any handle or width/height field. Smaller and Bigger uniformly scale by 10% around the center. Choose Stretch for independent dimensions; choosing Scale again restores the original aspect ratio. These edits retain original image bytes and are undoable.

Corners provides Square, Soft, Round and an exact radius field. Drag the small round handle inside a selected image or shape to adjust its corners. The eight square handles resize the element; the cursor indicates their direction.

Use V/T/R/F for select/text/rectangle/frame. Hold Shift to constrain proportions. O selects the Oval tool. Click Shapes in Insert or the toolbar for all figures. Pick one and click/drag on the page, or drag it directly from the picker. Select a figure and use Shape in the inspector to change its type while keeping its styling. Circle dimensions stay equal; Line has a thickness field. Alt resizes about the center or bypasses move snapping. Space+drag or the wheel pans. Cmd/Ctrl+wheel zooms; clicking the zoom percentage fits the page. On macOS, pinch the touchpad to zoom; two-finger scrolling pans. Pinch and modifier-scroll zoom stay anchored near the pointer. Cmd/Ctrl+S saves, Shift+Cmd/Ctrl+S saves as, O opens, N creates a blank document, Z undoes, Shift+Z redoes, D duplicates. B/I/U format selected text. Escape leaves text editing. Numeric inspector fields accept typed values; Enter commits and Escape cancels. Colors accept six-digit hex. Double-click a layer to rename it.

Clipboard currently handles text. Object duplication uses Cmd/Ctrl+D. Text is edited in independent boxes; automatic flow between pages, lists, tables, rotation, vector PDF/HTML export, DOCX import, collaboration and cloud features are later work. System fallback typography, screen-reader support and large-image performance need more validation. IME range handling is tested; all OS/language candidate window behavior is not yet certified.

Verify

cargo test --workspace
cargo run -p compose -- --smoke-test
cargo run -p compose -- --ui-smoke
cargo run -p compose -- --ui-smoke --refinements
cargo run -p compose -- --ui-smoke --polish
cargo run -p compose -- --ui-smoke --shapes --compact

The native smoke test sends events inside Compose; it doesn't move the real pointer or change the clipboard. It writes artifacts/native-smoke.compose. Use --ui-smoke-hold to leave its editable fixture open for visual inspection.

Run cargo run -p compose -- --ui-smoke --workflows for clipboard and replacement regressions. The checks leave your OS clipboard untouched. Object transfer between separate app instances remains unsupported; readable text is the fallback. See docs/FEATURE_GAPS.md for the prioritized remaining features. The refinement checks use artifacts/refinement-checks/, so the original sample is not overwritten. Add --show-text-selection with --ui-smoke-hold --refinements to inspect the selected-word toolbar. Native screenshots are saved as artifacts/editor-images-v2.png and artifacts/editor-text-v2.png.

Architecture is in ARCHITECTURE.md; design decisions and the original image references are in DESIGN.md and brand/. Font licenses and the small GPUI API bridge are documented under assets/fonts/ and vendor/README.md.

The 0.3 showcase and native screenshots are under artifacts/polish/. Use --ui-smoke-hold --polish --show-fonts or --show-colors for visual inspection, and --compact for the minimum window size. These tests stay inside Compose and do not move the system pointer or change the clipboard. Final visual QA uses the packaged application because macOS can defer presentation of fully occluded background windows.

Release limits: synchronized component instances, selection/group scaling, rotation, vector path editing, linked text pagination, further export/import formats and multi-platform release certification remain future milestones. This is a local macOS development build, not full Figma parity. Review availability is recorded in the validation artifacts.

Add images from your computer or the web (0.6)

Click Image in the toolbar or Insert area to open the native image picker. My computer opens the system file chooser. Search online searches the full web through Bing Images inside Compose: enter a phrase and press Enter, or use a suggestion. Commons remains a separate optional source. Select a thumbnail, inspect its title and originating website, and click Insert image. More images loads the next results page.

The selected original raster image is embedded in the .compose file; source, creator and license metadata, when provided, are saved with it and shown in the inspector. Search needs an internet connection; an inserted image stays available offline. Source and license links open only when clicked. Web search requires no API key; unknown usage rights remain unknown. If a website blocks its original download, Insert preview explicitly uses the cached lower-resolution thumbnail. Search errors also offer Open search in browser. Bing's public result markup can change; Compose reports provider failures instead of silently switching to a catalog.

Search and image decoding run on background workers. Closing the picker cancels pending UI effects; a newer search rejects older replies. The picker shows loading, empty, error/retry and preview-failure states. Supported originals are JPEG, PNG, WebP, GIF and BMP, limited to 64 MB / 32 megapixels. GIFs currently insert their first frame. Images retain the existing Scale/Stretch and corner controls.

Explicit integration checks (these access Bing and Commons):

cargo run -p image_search --example live
cargo run -p compose -- --image-picker-test
# Leave the results visible for inspection:
cargo run -p compose -- --image-picker-hold --compact

Offline tests use recorded search metadata. Live tests exercise native tab and query controls, thumbnails, original download, insertion, attribution, undo, reopening, local loading, invalid images and stale-request cancellation.

Export and print (0.7)

Click Export in the top bar, or press ⇧⌘E. Compose file exports a fully editable copy with every page and embedded image. The current document's save location and saved/unsaved state are retained. Save still updates the working file, and Save As changes its location.

Export PNG writes the current page without editor controls. Choose 1× (72 DPI), 2× (144 DPI), or 300 DPI; the menu shows the pixel dimensions. PNGs keep text layout, mixed fonts, image proportions, rounded corners, layer order, opacity, strokes and shadows. PNG export uses document coordinates and is independent of zoom or window size. Very large pages can require a lower resolution.

Print… or ⌘P prepares all document pages at 300 DPI and opens the macOS print dialog. Choose a printer, copies and page range, or use the dialog's Save as PDF. Cancelling leaves the document untouched. Print output currently uses raster pages; system paper/printer margins may scale the document. Native printing on Windows/Linux and vector PDF export remain future work.

Both operations run in background workers. They use local snapshots and never require a network connection. The File menu also exposes the export/print actions. The bundle includes the small native print adapter; users do not need Swift tools.

# Native export controls and a two-page print-to-PDF check (no physical printing):
cargo run -p compose -- --ui-smoke --output
# Also open and automatically dismiss the system print panel safely:
cargo run -p compose -- --ui-smoke --output --check-print-dialog

Select and organize layouts (0.8)

Use Shift-click on the canvas or in Layers to add/remove objects. Drag a rectangle from empty space to select several objects; hold Shift to add to the current selection. ⌘A selects page objects when you are not editing text. Drag the selection to move it together. Arrow keys nudge it by 1 pt; Shift+arrows use 10 pt. Copy/cut/paste, duplicate, delete and layer ordering operate on the whole selection and keep a single undo step.

The inspector shows selection bounds, six alignment controls and horizontal or vertical spacing for three or more objects. Group / ⌘G groups unlocked layers with the same parent. A normal click moves the group; double-click to edit a child. Escape returns to its group. Ungroup / ⇧⌘G restores the children and their positions. Groups follow their contents when children move or text reflows, preserve embedded images and reopen in .compose documents.

Grouping keeps selected layers' order and places the group at their frontmost selected layer. For nonadjacent layers, this can change their stacking relative to unselected objects between them. Grouping across different parents is currently disabled. Multiple-selection and group scaling are not yet implemented, so these selections show movement bounds without resize handles. Individual objects retain their normal resizing, corners and text formatting controls.

cargo run -p compose -- --ui-smoke --selection --compact
# Leave the real selection inspector visible:
cargo run -p compose -- --ui-smoke-hold --selection --compact
# Inspect the grouped version:
cargo run -p compose -- --ui-smoke-hold --selection --show-selection-group --compact

Touchpad zoom (0.8.1)

Pinch in or out over the workspace to zoom smoothly between 15% and 400%. Two-finger scrolling continues to pan, and Ctrl/⌘-scroll also zooms. The document point near the pointer stays fixed while zoom changes, including multiple input events before the next redraw. Sidebar/popover gestures do not zoom the page. Camera changes retain text/object selection, saved state and history.

# Tests the actual macOS view callback with synthetic AppKit gesture events:
cargo run -p compose -- --ui-smoke --touchpad --compact

License

Compose's original source code is MIT licensed. Vendored GPUI retains its Apache-2.0 license, and bundled fonts retain their included font licenses.

Word-compatible and Compose modes (0.9)

The Canvas / Word button in the toolbar opens the document mode chooser. Compose keeps the freeform editor: shapes, frames, groups, effects, image corners and arbitrary placement. Compose-mode documents cannot export DOCX; the option is disabled and the exporter enforces that restriction.

Word-compatible uses editable text and proportionate images in reading order with one-inch page margins. The inspector provides typography, text color, image width and block order controls. Shapes, groups, free placement, transparency and canvas effects are disabled. Export → Word document (.docx) produces real paragraphs, styled runs and embedded images, with page settings and section breaks. Font sizes, bold/italic/underline, colors, letter spacing, alignment and line spacing are retained. Word may substitute missing fonts and reflow content; exact pagination is not guaranteed. Install the bundled font families for closer output.

Both modes save editable .compose files, preserving the mode. Old files default to Compose. A compatible canvas can switch to Word mode and reflow; Undo restores its previous layout. A canvas using unsupported features cannot switch without losing them, so the chooser offers New Word document instead. New documents respect the normal unsaved-change prompt. File → New Word-compatible document… opens the creation screen with Word mode selected.

Word-mode blocks currently flow within explicit Compose pages; text is not linked between pages automatically. Oversized blocks show a warning. DOCX keeps their complete text for Word to paginate; PNG/Print rejects overflow instead of silently clipping it. Split oversized blocks across pages before using those outputs.

cargo run -p compose -- --ui-smoke --document-modes --compact
# Inspect the mode chooser and the real compatible editor:
cargo run -p compose -- --ui-smoke-hold --document-modes --compact

Create a document (0.11)

Opening Compose without a file shows the native creation screen. Choose Blank page, Résumé, Photo résumé, Photo cover, Report, Letter, Portfolio or Poster, select Compose or Word-compatible, name the document and click Create document. The preview renders the actual document. Every preset creates editable text and layers; The gallery scrolls to the second row at compact window sizes. Résumé works in both modes; Photo résumé, Photo cover, Portfolio and Poster require Compose to preserve their columns, placement, rounded images and shapes.

Open document… reopens a .compose file. Opening one from the command line goes directly to its editor. File → New document… / ⌘N returns to the creation screen. Choosing presets leaves your current document intact. Creating its replacement prompts to save unsaved changes; Back to document returns without replacing it. Both modes save locally as .compose.

In Compose, choose Text (T) and drag on the page to draw the box's width and height. A simple click inserts one default box. Shape and Frame use the same drag-to-size behavior, including reverse drags; Escape cancels before insertion. Typing retains your chosen height and expands it as needed. Single-click and drag to move an object; double-click to edit text. Drag the selected text border to move its box while editing, or right-click and choose Select / move text box. Word text retains its reading-order layout and fixed column width.

Right-click objects or Layers for working clipboard, duplicate, rename, order, lock/unlock and delete actions. Highlighted text keeps its range for copying. Locked objects can be inspected and unlocked through this menu, while dragging and editing stay blocked. The export menu explains which canvas features would be lost in Word; Compose-mode DOCX export stays disabled, so nothing is silently removed from your file.

cargo run -p compose -- --ui-smoke --creation --document-modes --compact

The Résumé contains editable name, contact details, profile, experience, education and skills. Photo résumé adds a portrait and two-column layout. Photo cover pairs a large image with an editorial title; Portfolio adds project details and a large work sample. Their photographs are normal image layers: select one and choose Replace image… to use a local file or web search. The original bytes, proportions and editable corners survive .compose reopening. No network access is needed to use a preset. Generated sample images and their prompts are documented in assets/presets/SOURCES.md.

cargo run -p compose -- --ui-smoke --presets --compact
# Inspect the native gallery or editable portrait résumé:
cargo run -p compose -- --ui-smoke-hold --presets --compact
cargo run -p compose -- --ui-smoke-hold --presets --show-photo-editor

Drag, hyperlinks and background removal (0.12)

Drag rows in Layers to reorder them. The blue edge shows the drop location. Canvas stacking changes without moving objects; Word mode reflows its reading order. Layers may move between existing groups by dropping beside a layer in the destination group; invalid nesting and locked content are rejected.

Drag Text, Shape or Frame from Insert, or a component from Assets, onto the page. Dragging Image opens its picker at that position. Drop image files from Finder/Explorer directly onto the page; up to 16 are decoded in the background and inserted as one undo operation. Dropping a .compose file opens it with the normal unsaved-change protection.

Highlight text and press Ctrl/⌘K, or use the link icon, right-click menu or inspector. Apply a web or email address, edit/remove it, or open it. Ctrl/⌘-click follows links in the editor; Preview uses a normal click. Links retain their range through formatting, clipboard and .compose saves, and export as clickable Word hyperlinks. PNG and the current raster print/PDF output preserve their appearance but cannot carry interactive links.

Remove background in the image inspector and Without background in image search now use one shared Compose service. Users need no account or API key. The desktop app contains only the public service address; the provider key stays on the server. The selected image is uploaded when requested, and a transparent preview PNG is returned. Shared usage limits apply. Position/style, original pixels and Undo are preserved; failed or stale replies keep the working image. There is no desktop ML runtime, model download or GPU inference requirement.

The server uses bgclear's preview API with size=preview (up to 800 px); paid full-resolution size is never requested. Deployment files and operator instructions are in services/background/README.md. The service is a separate Rust executable with bounded requests, concurrency and shared limits; it stores no images and has no user accounts/database.

The macOS and Windows packaging scripts include the live shared service at https://compose-background.onrender.com/v1/background/remove by default. Builders can override the public COMPOSE_BACKGROUND_SERVICE_URL before packaging. Normal users do not configure anything. A local runtime override can point to http://127.0.0.1:8787/v1/background/remove during development. Builds without an endpoint explain that the service is not connected yet. The local pipeline is tested with a real relay and controlled provider responses. The Render deployment and packaged native Remove background button were also verified against the live provider on 2026-10-11, including transparent PNG output, unchanged geometry, Undo/Redo and editable save/reopen. The provider key stays in the server's secret settings. The root render.yaml creates a Render Free Docker service in Frankfurt. Render Free can sleep after 15 idle minutes; Compose checks health before uploading and allows up to two minutes for the service to wake.

Windows development needs Rust's MSVC toolchain and Visual Studio C++ Build Tools with the Windows SDK (GPUI compiles its DirectX shaders using fxc.exe). Run ./scripts/package-windows.ps1 for an optimized portable Windows ZIP. The native desktop GitHub workflow builds the Windows ZIP, checks its extracted executable and uploads the archive with its SHA-256 checksum. Windows GUI/hardware QA is still pending. Deployment instructions are in website/README.md.

cargo run -p compose -- --ui-smoke --actions --compact