Fixed macOS window screenshots across apps and displays — screenshot --window now uses the native macOS window capture path, which correctly captures apps like Finder, Preview, Slack, and Chrome instead of depending on fragile CoreGraphics window capture behavior.

usecomputer window list --json
usecomputer screenshot ./finder.png --window 51448 --json

Fixed region crops on secondary displays — region screenshots now translate desktop-global coordinates into display-local coordinates before cropping, so --display + --region captures the right area on multi-monitor setups.

usecomputer screenshot ./region.png --display 1 --region "100,100,800,600" --json

Added repeatable --modifier <key> support to click — hold one or more keyboard modifiers while clicking, using the same modifier names and aliases accepted by press:

usecomputer click -x 600 -y 400 --modifier option
usecomputer click -x 600 -y 400 --modifier cmd --modifier shift

usecomputer click --help now shows command-specific help with examples for this flow.

Fixed negative -x and -y values in CLI commands — commands that accept optional coordinate flags now correctly parse negative screen positions for multi-display layouts:

usecomputer hover -x -1200 -y 300
usecomputer debug-point -x -800 -y 240

Fixed GitHub release archives for standalone and C API downloads — CI now always uploads release assets when a matching GitHub release exists, and Windows builds include usecomputer_c.dll from the correct output path.

C API shared library — libusecomputer_c is now built for all platforms and included in GitHub releases. Use from any language with FFI (Julia, Python ctypes, Ruby FFI, etc.):

#include "usecomputer.h"

double x, y;
uc_mouse_position(&x, &y);

char* displays = uc_display_list();
char* result = uc_screenshot(NULL, -1, -1);
if (!result) fprintf(stderr, "%s\n", uc_last_error());
else { puts(result); uc_free(result); }

Download from this release — each archive contains the standalone CLI, N-API .node, C shared library (libusecomputer_c.dylib/.so/.dll), and usecomputer.h header.

All releases now include the standalone executable — each platform archive contains usecomputer (or usecomputer.exe on Windows), a self-contained binary with no Node.js dependency.

Fixed usecomputer command on Windows — the npm bin entry now works from CMD and PowerShell without needing Git Bash or MSYS2 in PATH. The launcher is now a compiled TypeScript file (dist/bin.js) instead of a shell script:

npm install -g usecomputer
usecomputer screenshot ./shot.png --json

Removed clipboard get and clipboard set commands — these were only implemented on Windows and returned NOT_SUPPORTED on macOS and Linux. Use keyboard shortcuts instead:

usecomputer press "cmd+c"   # copy (macOS)
usecomputer press "ctrl+c"  # copy (Windows/Linux)
usecomputer press "cmd+v"   # paste (macOS)
usecomputer press "ctrl+v"  # paste (Windows/Linux)

Kitty Graphics Protocol support — screenshot now emits the PNG inline to stdout using the Kitty Graphics Protocol. Set AGENT_GRAPHICS=kitty and the image lands directly in the AI model's context window — no separate file-read step needed:

AGENT_GRAPHICS=kitty usecomputer screenshot ./shot.png --json
# { ..., "agentGraphics": true }

Works with kitty-graphics-agent, an OpenCode plugin that intercepts the escape sequences and injects them as LLM-visible image attachments. Add to opencode.json:

{ "plugin": ["kitty-graphics-agent"] }

Aligned table output for list commands — display list, window list, and desktop list now render as aligned, human-readable tables. JSON mode (--json) is unchanged.

Fixed agentGraphics JSON field — the field now reflects actual Kitty emission success. Empty PNG files and I/O errors correctly report false.

Standalone executable — ships as a self-contained binary, no Node.js required at runtime

Linux X11 screenshot support — capture screens on Linux via XShm with XGetImage fallback on XWayland

Screenshot coord-map and scaling — longest edge scaled to 1568 px with coordMap field for accurate pointer remapping:

usecomputer screenshot ./shot.png --json
usecomputer click -x 400 -y 220 --coord-map "0,0,1600,900,1568,882"

New debug-point command — draw a red marker at a mapped coordinate to validate clicks before executing:

usecomputer debug-point -x 400 -y 220 --coord-map "0,0,1600,900,1568,882"

Keyboard synthesis — type and press commands for text input and key chords:

usecomputer type "hello from usecomputer"
usecomputer press "cmd+s"
cat ./notes.txt | usecomputer type --stdin --chunk-size 4000

Native scroll support — scroll in any direction at any screen position:

usecomputer scroll --direction down --amount 5

Library exports — import as a Node.js library for use in custom agent harnesses:

import * as usecomputer from 'usecomputer'
const shot = await usecomputer.screenshot({ path: './shot.png', display: null, window: null, region: null, annotate: null })

OpenAI and Anthropic computer-use examples — README includes full agentic loop examples for both providers