Scale computer-use 2.0 with open-source drivers, cross-OS sandboxes, and benchmarks for training, evaluation, and data generation.
Rust agentai-agentapplecomputer-usecomputer-use-agentcontainerizationcuadesktop-automationhacktoberfestlumemacosmanusoperatorswiftvirtualizationvirtualization-frameworkwindowswindows-sandbox
Examples
| Example | What it is |
|---|---|
cua-bots-macos |
Cua Bots for macOS: persistent bots, each with a koala face, its own Space, a memory in the Cua Volume layout, routines, approvals, custom rules and saved sign-ins. SwiftUI on the Cua Swift SDK and the Cua Spaces app export. The featured example; source-available (FSL-1.1-MIT) |
cua-bots-ios |
Cua Bots for iPhone: a remote client for the Mac's bots; chat, watch and take over their computers, answer approvals, get notifications. Source-available (FSL-1.1-MIT) |
openkoalabot-example-swift |
OpenKoalaBots, a chat app for agent coworkers on Cua Spaces: a SwiftUI app on the Cua package via the Swift overlay libs/spaces-sdk-swift and CuaSpacesStreaming (libs/spaces-app-swift). The first of the three; its 209-test suite and RUBRIC.md are the behaviour spec. Source-available (FSL-1.1-MIT) |
openkoalabot-example-tauri |
The same app as a Rust core on the cua-spaces crate, with a Tauri 2 + React shell |
openkoalabot-example-ts |
The same app in TypeScript on @trycua/cua: a Node agent loop, a local server, and a web UI that streams through @trycua/cua/browser (gRPC-Web + the ticketed /media socket) and WebCodecs |
openkoalabot-example-scenario |
The shared, language-neutral scenario all three run headlessly: add or claim a Space, stream, run an agent thread (fake CLI), send a file, teleport a generated profile, run presence with two clients, release |
infinite-canvas-swift |
infinite-canvas: every window of every Space on one zoomable canvas, with agent and coworker cursors and agent threads. SwiftUI and AppKit on the Cua Swift packages. Source-available (FSL-1.1-MIT) |
driver |
cua-driver samples |
python |
Python samples |
agents-in-sandboxes |
Run a coding-agent harness in a sandbox image over a task file, then collect verified results into a JSON report |
streaming |
The same streaming scenario against a cua-spacesd in Python, Rust, Swift, Kotlin and TypeScript (Node and browser); see SCENARIO.md |
Run all three OpenKoalaBots examples against the same Space kinds:
examples/openkoalabot-example-scenario/run.sh --impl all --lane fixture # hermetic
examples/openkoalabot-example-scenario/run.sh --impl all --lane docker # linux under gVisor, one container at a time
Each implementation's README has its build, test and run commands.