Open-source alternatives to Computer Use Preview

Page-agent is an LLM browser automation agent and JavaScript in-page GUI controller. It translates natural language instructions into direct browser interface actions to automate web-based tasks and manipulate web page elements through a programmable interface. The system coordinates complex sequences of actions across multiple browser tabs and different websites. It functions as a remote browser control server, providing an interface that allows external clients to operate a browser and manage page interactions. Its capabilities include natural language intent decoding and action mapping, D

LaVague is an LLM web agent framework and large action model designed to translate natural language instructions into executable browser automation scripts. It functions as a multi-modal orchestrator that reasons over web page states and HTML content to automate multi-step tasks via a Selenium-based automation engine. The framework features a modular model provider layer, allowing users to swap between different language and vision models from providers such as Anthropic, Gemini, and Azure OpenAI. It employs a multi-modal world model to process screenshots and HTML structures, utilizing retri

This project is an automation framework that connects large language models to web browsers via the Chrome DevTools Protocol for autonomous task execution. It functions as a bridge between intelligent agents and browser engines, allowing for the direct control of browser sessions and profiles. The framework features a self-healing agent capable of generating and executing custom scripts during runtime to resolve failures and optimize browser tasks. It supports stealthy deployment through the use of integrated proxies and captcha solvers to bypass bot detection and security mitigations. The s

This project is an LLM browser automation framework and AI agent browser interface. It serves as a control layer that translates natural language instructions into browser interactions using large language models, enabling AI agents to navigate and interact with web pages through standardized browser-control functions. The system functions as an RPA workflow orchestrator and headless browser management tool, capable of recording and replaying deterministic browser sequences to automate repetitive tasks. It distinguishes itself through stealth configurations, including residential proxies and

Mastra is an orchestration framework designed for building, deploying, and managing autonomous AI agents and multi-agent systems. It provides a comprehensive suite of primitives for creating resilient AI applications, including durable workflow orchestration, event-driven agent loops, and semantic memory management. By integrating these core components, the platform enables developers to build complex, multi-step processes that can reason about goals and execute tasks without manual intervention. The framework distinguishes itself through its focus on observability and secure, isolated execut

Microsandbox is a runtime for creating and managing lightweight, hardware-isolated virtual machines — called sandboxes — that boot directly from standard OCI container images. Each sandbox runs as its own host process with a separate kernel, filesystem, and network stack, providing process-per-sandbox isolation. The project includes a command-line tool and multi-language SDKs (Rust, TypeScript, Python, Go) for programmatic lifecycle control, and it communicates with sandbox agents over Unix sockets using a CBOR-encoded protocol. What distinguishes Microsandbox is its combination of host-manag

Osmedeus is a security workflow orchestration engine that coordinates AI agents, shell commands, and scanning tools through declarative YAML pipelines. It functions as a distributed security scanner, a declarative workflow automator, and an AI agent framework for security, enabling automated multi-step security analysis with conditional branching, parallel execution, and distributed workers. The engine distinguishes itself through a hybrid runner model that executes workflow steps on the local host, inside Docker containers, or over SSH to remote machines, selected per step or module. It supp

Cronicle is a distributed job scheduler that replaces traditional cron with a browser-based management interface. It runs scheduled tasks across a cluster of servers with automatic failover, using a custom cron parser that intersects day-of-month and day-of-week constraints when both are specified. The system executes jobs through a plugin framework that runs command-line scripts in any language, communicating via JSON over standard input and output. The scheduler provides a web-based real-time dashboard for monitoring running jobs with live logs, resource usage charts, and progress updates.

Bard-API is an asynchronous Python wrapper and client for interacting with Google Gemini. It functions as a stateful conversation manager and multimodal interface, allowing users to send text and image prompts to a language model and retrieve responses. The library utilizes a cookie-based authentication system that extracts session tokens from local browser storage to authorize requests. To manage access and connectivity, it includes proxy-based request routing to bypass regional restrictions and avoid IP blocks. The project covers capabilities for multimodal AI analysis and the maintenance

how2 is a terminal-based tool that translates plain-English questions into shell commands using AI and StackOverflow data. It functions as a command-line interface where users describe what they want to do in natural language, and the tool returns the appropriate Unix shell or PowerShell command, with support for generating multi-line Bash scripts from natural language prompts. The tool distinguishes itself through its interactive answer browsing mode, which lets users select and copy from multiple StackOverflow answers directly in the terminal. It includes a fallback search mechanism that qu

kubectl-ai is a natural language cluster operator and AI command assistant that translates plain-text prompts into executable Kubernetes commands. It serves as an interface between large language models and the Kubernetes API to enable cluster management through conversational text. The project implements a Model Context Protocol server to expose cluster operations as standardized tools for external AI clients. It uses a provider-agnostic model interface to support both cloud-based and local AI backends. The system covers natural language infrastructure control and AI-assisted DevOps through

gstack is an AI agent framework and development workflow system designed to automate the software development lifecycle. It coordinates specialized AI personas to manage tasks across product design, engineering management, and quality assurance, transforming product intent into technical specifications and final releases. The project is distinguished by its deep integration of headless browser automation and semantic code memory. It utilizes a persistent Chromium daemon for web scraping and visual auditing, and implements a searchable knowledge base that logs architectural decisions and repos

Droidrun is a mobile device automation framework that uses large language models to translate natural language commands into executable actions on mobile operating systems. It functions as an agent orchestrator and UI automation engine, providing a reasoning engine that decomposes complex mobile tasks into smaller, manageable steps. The system distinguishes itself through a hierarchical action translation process and the ability to analyze accessibility trees and screenshots to determine the visual layout and current status of mobile applications. It supports execution across both physical ha

Nanobrowser is an AI browser automation tool and Chrome extension that uses large language models to execute complex, multi-step web workflows through a natural language interface. It functions as a multi-agent workflow orchestrator, coordinating specialized AI agents to plan strategies and interact with page elements to complete tasks. The system emphasizes local-first operations, acting as a local API manager that stores provider credentials and executes data processing within the browser to keep sensitive information and keys out of external servers. It utilizes a provider-agnostic API bri

This project is a Model Context Protocol server that bridges artificial intelligence agents with cloud-based web scraping and automation resources. It functions as a remote task orchestrator, allowing agents to discover, configure, and execute complex browser automation jobs as callable functions within their native environments. The server distinguishes itself by providing a unified framework for managing distributed workflows, including the ability to handle asynchronous task polling, structured data serialization, and real-time status tracking. It supports advanced agentic capabilities suc

The Gemini Cookbook is a comprehensive collection of implementation patterns, code samples, and development guides designed for building applications with Google Gemini models. It serves as a central resource for developers to integrate multimodal generative artificial intelligence into their software, providing the necessary frameworks to manage model interactions, stateful workflows, and structured data extraction. The repository distinguishes itself by offering specialized toolkits for autonomous agent orchestration, enabling the construction of agents that can execute code, browse the web

This project is an LLM-powered web crawler and data extractor that uses large language models to navigate websites and parse content into structured JSON or Markdown formats. It functions as an automated browser orchestrator and domain discovery engine, interpreting plain English instructions to identify relevant pages and extract specific information. The system distinguishes itself through agentic browser automation, allowing it to perform human-like interactions such as clicking buttons and scrolling based on natural language commands. It employs goal-oriented crawling to analyze website s

Shell-ai is a command-line interface tool that translates plain English text into executable shell commands using language models. It acts as a terminal assistant that bridges natural language prompts and operating system shell execution to generate, review, and run system commands. The application incorporates a model-agnostic endpoint abstraction that delegates generation requests across multiple external language model providers and API endpoints through a unified interface. It features a config-driven provider selection mechanism that maps configuration files and environment variables dyn

Jasper Client is a voice computing client and extensible speech framework designed to translate natural language speech into hardware actions and service requests. It functions as a voice command interface that manages the end-to-end process of audio capture, transcription, and action execution. The system features a modular architecture that allows for the integration of custom plugins, various speech recognition engines, and synthesis providers. This plugin-based approach supports the addition of new speakers and regional language capabilities without altering the core logic. The client in

This project is an automation suite comprising an AI visual asset generator, a browser-based social publisher, an Electron resource extractor, and a Markdown content transformer. It functions as a content automation pipeline that uses large language models to generate text and images for distribution across social media platforms. The system distinguishes itself through specialized visual generation capabilities, producing professional infographics, slide decks, educational comics, and SVG diagrams via structured prompts. It also features a dedicated workflow for extracting resources from Ele

01 is a voice-to-code agent and language model voice interface framework that enables natural language control of computers and devices. It functions as a real-time audio streaming server and a cross-platform voice client, translating spoken instructions into executable code to automate software, manage files, and browse the web. The system supports both local and cloud-based language models, alongside local or hosted speech-to-text and text-to-speech engines. It is designed for custom hardware integration, providing the means to build embedded AI voice controllers using microcontrollers like

Firecrawl is a headless browser automation tool and web crawling engine designed to extract structured data from the web. It functions as an API that transforms raw website content and documents into clean markdown and JSON formats to serve as context for large language models. The project distinguishes itself by using natural language prompts to translate human instructions into targeted data extraction tasks and browser actions. It can execute interactive page navigation, such as clicking and scrolling, and perform automated web research to retrieve structured data without manual interventi

Magentic-UI is an agentic UI toolkit and framework that enables large language models to interface with real-time browser environments, operating systems, and virtual machines. It provides a sandbox environment where models can execute instructions to manage local files and run shell commands. The project functions as a web interaction orchestrator and browser automation framework, allowing for the execution of end-to-end web workflows and form completions. It coordinates these actions through a system that translates natural language goals into executable sequences. The toolkit covers sever

This project is a Model Context Protocol server that enables Large Language Models to control Playwright browsers for web automation, scraping, and end-to-end testing. It functions as a programmable interface for executing JavaScript, capturing screenshots, and interacting with web elements across multiple browser engines. The server exposes browser automation capabilities as a set of standardized tools that models can discover and invoke. It supports session-based browser isolation to ensure unique contexts for each client connection and provides a transport layer using either standard input

BrowserMCP is a browser automation bridge that connects AI tools to a live browser session through a local proxy server. It implements a standardized protocol for sending commands like click, type, and navigate to a real browser instance running on the user's machine, while keeping all browsing data on the device. The project distinguishes itself by preserving user sessions and fingerprints across automation tasks. It attaches to the user's existing browser profile to maintain cookies, logins, and authentication state, and uses the real browser's user agent, viewport, and extension context to

AICommand is an LLM-powered editor controller and automation tool designed for the Unity Editor. It integrates large language models to translate natural language prompts into executable commands, allowing users to manipulate development environment settings and objects through text-based instructions. The system employs a text-to-action pipeline that uses LLM-based command mapping and reflection-based API discovery to resolve natural language input into specific editor functions. This process enables the execution of Unity editor functions by resolving string identifiers to code handles at r