Open-source OpenAI Codex alternatives

OpenCode is an open-source coding agent that helps developers write, change and understand code using AI models. It is built in TypeScript, released under the MIT license, and installed through an install script or, in beta, as a desktop application for macOS, Windows and Linux.

It ships with two built-in agents that you switch between with the Tab key. The build agent has full access for everyday development work, while the plan agent is read-only: it denies file edits by default and asks permission before running shell commands, which suits exploring an unfamiliar codebase or planning a change. A general subagent handles complex searches and multistep tasks and can be called with @general in messages.

Configuration is covered in the project documentation, and the install location can be controlled with environment variables. OpenCode suits developers who prefer working from the command line and want an agent whose source code they can read and modify.

Gemini CLI is Google's open-source terminal agent for working with Gemini models from the command line. It offers a lightweight path from a prompt on the command line to the model, and it is designed for developers who spend their day in a shell. The project is written in TypeScript and licensed under Apache-2.0.

Built-in tools cover Google Search grounding, file operations, shell commands and web fetching, and further integrations can be added through the Model Context Protocol (MCP). The project highlights access to Gemini models with a large context window, and a free tier when you sign in with a personal Google account.

You can run it instantly with npx or install it globally with npm, Homebrew, MacPorts or Anaconda. It has preview, stable and nightly release channels published on a regular schedule. It suits developers who want an AI assistant for code and shell tasks without leaving the terminal.

OpenHands is an open-source project for AI-driven software development. Its current README centers on Agent Canvas, a self-hosted developer control center for starting conversations with coding agents and automating everyday engineering tasks. It is written mainly in TypeScript and released under the MIT license.

By default Agent Canvas uses the open-source OpenHands agent, but it can also drive third-party agents such as Claude Code, Codex and Gemini, or any ACP-compatible agent. It runs locally by default and can connect to several agent backends, including Docker containers, virtual machines and company infrastructure, so agents can be moved between local, remote and cloud setups without interrupting your work.

Automations can be triggered on a timetable or by incoming webhook events and connect with tools such as Slack, GitHub, Linear and Notion, for example publishing reports to Slack or breaking GitHub issues into tasks. Agents can optionally run on OpenHands Cloud or OpenHands Enterprise. It suits engineering teams that want to coordinate several coding agents from one self-managed place.

Cline is an open-source coding agent that works in your IDE, terminal and desktop. It can edit code across a project, create files, run commands and browse the web, with human-in-the-loop approval for tools in the editor. It is written in TypeScript and released under the Apache-2.0 license.

The same agent engine is offered in several forms: a CLI with an interactive mode and a headless mode for CI/CD and scripting, a native desktop app for macOS and Windows that works from any folder, schedules recurring routines and manages models, plugins and MCP servers, a VS Code extension, a JetBrains plugin for IntelliJ IDEA, PyCharm, WebStorm, GoLand and others, and an SDK.

The SDK is a Node.js programmatic API for building custom agents with custom tools, multi-agent teams, connectors and scheduled automations. The README notes that the JetBrains plugin is not open-sourced. Cline suits developers who want one coding agent across their editor, terminal and automation pipelines.

goose is an AI agent for general tasks that runs locally on your own machine. It is not limited to coding: the project suggests using it for research, writing, automation, data analysis and similar tasks. It is built in Rust, released under the Apache-2.0 license, and available as a desktop application on Windows, Linux and macOS, a full CLI for terminal workflows, and an API for embedding it elsewhere.

goose works with more than a dozen model providers, including Anthropic, OpenAI, Google, Ollama, OpenRouter, Azure and Bedrock, and can use existing Claude, ChatGPT or Gemini subscriptions through ACP as an alternative to API keys. It connects to many extensions through the Model Context Protocol open standard, and custom distributions let teams ship their own goose build with preconfigured providers, extensions and branding.

The project is part of the Agentic AI Foundation at the Linux Foundation, and its documentation covers a quickstart, installation, tutorials and governance. It suits developers and knowledge workers who want a configurable local agent that is not tied to a single model vendor.

Kilo Code is an AI coding agent that you can use in VS Code, in JetBrains IDEs or from the CLI. It is open source under the MIT license and written in TypeScript. According to the README, you can pick from more than 500 models and switch between them in the middle of a task, and you pay the model provider's rate with no markup. No API keys are needed to start.

Besides the local extensions, a Cloud Agent runs Kilo from the web without a local machine, and automated AI code reviews can be set up for pull requests. The CLI is started with the kilo command in a project directory and can be installed from release binaries for Windows, macOS on Apple Silicon and Intel, and Linux on x64 and ARM, with variants for older CPUs and Alpine-style images. The repository started in 2025 and its README is translated into many languages.

Aider is AI pair programming in your terminal. It lets you work with large language models to start a new project or build on an existing codebase, asking for changes in plain language and letting the model edit the files in your repository.

It builds a map of the whole codebase to help in larger projects, supports more than 100 programming languages and commits changes to git with descriptive messages, so you can diff or undo AI edits with familiar tools. It can connect to cloud models and local models, accept images and web pages as context, take voice instructions, and run linters and tests after each change, attempting to fix problems they report. It can also be used from an IDE through code comments, or via copy and paste with a model's web chat.

Aider is written in Python and licensed under Apache-2.0. The README names a handful of commercial models it has been used with most, but says it can connect to almost any model, and installation and usage guides are available in the documentation.