# cua | Cua | Knowledge Base | Instagit

Open-source infrastructure for Computer-Use Agents. Sandboxes, SDKs, and benchmarks to train and evaluate AI agents that can control full desktops (macOS, Linux, Windows).

GitHub Stars: 14.6k

Repository: https://github.com/trycua/cua

---

## Articles

### [How to Contribute Custom Environment Adapters to the Cua-Bench Registry](/trycua/cua/contribute-custom-environment-adapters-cua-bench-registry)

Learn how to contribute custom environment adapters to the Cua-Bench registry. Implement an EnvironmentAdapter class, define task metadata, generate the registry, and submit a pull request.

- Tags: how-to-guide
- Published: 2026-04-27

### [How Cua Handles Screen State Verification and Element Detection for Reliable Automation](/trycua/cua/cua-screen-state-verification-element-detection-automation)

Learn how Cua achieves reliable automation by combining AI screenshot validation with dual-path element discovery using accessibility trees and computer vision OCR.

- Tags: how-to-guide
- Published: 2026-04-27

### [How to Implement Rate Limiting and Cost Controls for LLM API Calls in Cua](/trycua/cua/implement-rate-limiting-cost-controls-llm-api-calls-cua)

Implement LLM API rate limiting and cost controls in Cua. Learn to use exponential backoff retries and BudgetManagerCallback for budget management and agent execution stops.

- Tags: how-to-guide
- Published: 2026-04-27

### [How to Scale Cua Agents Across Multiple Cloud VMs: Async Fleet Management Guide](/trycua/cua/scaling-cua-agents-multiple-cloud-vms)

Scale Cua agents across multiple cloud VMs efficiently. Learn how to use VMProviderType.CLOUD and asyncio.Semaphore for concurrent agent execution while managing API limits and resources.

- Tags: how-to-guide
- Published: 2026-04-27

### [How Cua's MCP Server Session Management Works for Long-Running Agent Tasks](/trycua/cua/cua-mcp-server-session-management-long-running-agents)

Discover how Cua's MCP server uses session management and idle detection to efficiently handle long-running agent tasks without resource waste.

- Tags: internals
- Published: 2026-04-27

### [How to Export Agent Trajectories for Training Reinforcement Learning Models in Cua](/trycua/cua/export-agent-trajectories-reinforcement-learning-models)

Export Cua agent trajectories for RL training using the CLI or zip_trajectory() function. Streamline your reinforcement learning pipelines with this essential data format. Learn how now.

- Tags: how-to-guide
- Published: 2026-04-27

### [cua-computer vs cua-computer-server: Understanding the Client-Server Architecture](/trycua/cua/difference-cua-computer-cua-computer-server-packages)

Learn the difference between cua-computer client SDK and cua-computer-server. Understand how this client-server architecture enables remote execution and sandbox management.

- Tags: deep-dive
- Published: 2026-04-27

### [How Cua Handles Multi-Touch Gestures and Mobile-Specific Interactions](/trycua/cua/cua-handle-multi-touch-gestures-mobile-interactions)

Discover how Cua handles multi-touch gestures and mobile interactions with its transport-agnostic architecture and high-level async API. Supports ADB and gRPC.

- Tags: internals
- Published: 2026-04-27

### [How to Create a Custom Skill for CuaBot to Automate Specific GUI Tasks](/trycua/cua/create-custom-skill-cuabot-gui-automation)

Learn to create a custom skill for CuaBot to automate GUI tasks. Follow our guide or use the CLI to record and train your unique automation workflows.

- Tags: how-to-guide
- Published: 2026-04-27

### [Security Model for Running Untrusted Agent Code in Cua Sandboxes](/trycua/cua/security-model-untrusted-agent-code-cua-sandboxes)

Discover the Cua security model for running untrusted agent code. Learn how Cua uses sandboxing with process isolation, privilege restriction, and network sandboxing for robust defense.

- Tags: architecture
- Published: 2026-04-27

### [How to Debug Cua Agent Failures Using Telemetry and Tracing: A Complete Guide](/trycua/cua/debug-cua-agent-failures-telemetry-tracing)

Debug Cua agent failures effectively. Learn to enable telemetry and tracing for detailed crash analytics, session logs, screenshots, and API call sequences to pinpoint and resolve issues.

- Tags: how-to-guide
- Published: 2026-04-27

### [How Lume Uses Apple's Virtualization Framework for Near-Native macOS Performance](/trycua/cua/lume-apple-virtualization-framework-macos-performance)

Discover how Lume uses Apple's Virtualization Framework to achieve near-native macOS performance on Apple Silicon. Run macOS VMs directly with hardware acceleration and paravirtualized I/O.

- Tags: how-to-guide
- Published: 2026-04-27

### [Omni vs UI‑TARS: Differences Between VLM Model Options (omni, uitars, uitars‑mlx, uitars‑hf) in CUA](/trycua/cua/differences-vlm-model-options-omni-uitars-uitars-mlx-uitars-hf)

Explore the differences between CUA's VLM model options including OmniParser and UI-TARS variants (uitars, uitars-mlx, uitars-hf). Choose the best fit for your hardware & privacy needs.

- Tags: deep-dive
- Published: 2026-04-27

### [How to Add a Custom Benchmark Task to Cua-Bench: Complete Implementation Guide](/trycua/cua/add-custom-benchmark-task-cua-bench)

Learn how to add a custom benchmark task to Cua-Bench. Follow this implementation guide to create and validate your own tasks for the Cua-Bench framework.

- Tags: how-to-guide
- Published: 2026-04-27

### [How Cua-Bench Evaluates Agents on OSWorld, ScreenSpot, and Windows Arena Benchmarks](/trycua/cua/cua-bench-evaluate-agents-osworld-screenspot-windows-arena)

Cua-Bench evaluates agents on OSWorld, ScreenSpot, and Windows Arena benchmarks using specialized sandboxes and advanced verification methods. Learn how it computes success metrics.

- Tags: how-to-guide
- Published: 2026-04-27

### [How to Integrate Cua with Claude Code or Cursor Using the MCP Server](/trycua/cua/integrate-cua-claude-code-cursor-mcp-server)

Integrate Cua with Claude Code or Cursor using the MCP server. Start the MCP server with `cua serve-mcp` and register it with `claude mcp add cua` for seamless sandbox management and computer control.

- Tags: how-to-guide
- Published: 2026-04-27

### [Performance Overhead of Running Cua Agents in Sandboxes vs Native Execution](/trycua/cua/cua-agent-sandbox-vs-native-execution-performance-overhead)

Discover the performance overhead of running Cua agents in sandboxes versus native execution. Understand cold-start latency and serialization costs for optimized performance.

- Tags: performance
- Published: 2026-04-27

### [How Cua Driver Achieves Background macOS Automation Without Stealing Cursor or Focus](/trycua/cua/cua-driver-background-macos-automation-cursor-focus)

Discover how Cua Driver enables background macOS automation by preventing cursor and focus stealing using accessibility, synthetic focus, and reactive workspace notifications.

- Tags: internals
- Published: 2026-04-27

### [How to Implement Custom Action Spaces for Specific Automation Tasks in Cua](/trycua/cua/implement-custom-action-spaces-cua-automation)

Learn to implement custom action spaces in Cua for automation tasks. Define function signatures, inject into prompts, parse responses, and map actions to sandbox commands.

- Tags: how-to-guide
- Published: 2026-04-27

### [Cua Agent Reasoning Loop Architecture: How It Handles Tool Failures](/trycua/cua/cua-agent-reasoning-loop-architecture-tool-failures)

Explore the Cua Agent reasoning loop architecture. Discover how it converts LLM responses to commands and handles tool failures by degrading calls to prevent crashes.

- Tags: architecture
- Published: 2026-04-27

### [How Cua's Trajectory Recording and Replay System Works for Agent Debugging](/trycua/cua/cua-trajectory-recording-replay-system-agent-debugging)

Debug agents effectively with Cua's trajectory recording and replay system. See how Cua captures and replays agent actions for clear visualization.

- Tags: internals
- Published: 2026-04-27

### [How to Configure Custom Sandbox Images for Linux, macOS, Windows, and Android in Cua](/trycua/cua/configure-custom-sandbox-images-for-os-environments)

Learn to configure custom sandbox images for Linux macOS Windows and Android using Cua's immutable Image builder. Effortlessly spin up specific OS environments for your testing needs.

- Tags: how-to-guide
- Published: 2026-04-27

### [Cloud vs Local Sandbox Execution in Cua: Key Differences and When to Use Each](/trycua/cua/cua-cloud-vs-local-sandbox-execution-modes)

Understand Cua's cloud vs local sandbox execution modes. Learn how remote VMs and local runtime differ and when to use each for your Cua projects.

- Tags: deep-dive
- Published: 2026-04-27

### [How Cua's Multi-Provider VLM Architecture Supports OpenAI, Anthropic, Gemini, Qwen, and UITARS](/trycua/cua/how-does-cua-multi-provider-vlm-architecture-support-openai-anthropic-gemini-qwen-local-models)

Discover how Cua's multi-provider VLM architecture seamlessly integrates OpenAI, Anthropic, Gemini, Qwen, and local models like UITARS using a unified OpenAI-compatible format. Switch providers effortlessly.

- Tags: architecture
- Published: 2026-04-27