What it is
Ever opened a project full of unfamiliar code and wondered where on earth to begin? OpenCode is an AI coding assistant that can look through a project with you, explain how it works, suggest improvements and help make changes. You ask questions in ordinary language. It works with the actual files, so you do not have to copy and paste every snippet into a chat window.
The slightly technical name for this is a coding agent. Unlike a chatbot that only gives advice, an agent can use tools: it can inspect files, run commands and edit code. That’s useful, but it also means its actions deserve your attention. OpenCode is available as a desktop app for Windows, macOS and Linux, or as a terminal interface (the text-based window developers use to run commands).
OpenCode has two helpful ways of working: Plan is for examining a problem and discussing what to do; Build is for actually making changes. If you are learning, start with Plan and ask it to explain a small project before inviting it to edit anything.
The software itself is open source and free. It can work with a choice of AI models, including free options; connecting a different provider may involve charges. You do not need to pick a complicated setup just to try it. The official product page explains the available apps and model choices.
What’s new
OpenCode moves quickly, so dates matter. The most recent main release I verified was version 2.3.0, released on 13 November 2026. It makes working on several tasks feel less like watching one long conversation scroll past.
Helpers can work in the background. OpenCode calls these subagents: smaller AI helpers working on parts of a larger job. They can now continue while the main conversation remains available, then report back when they finish. You can keep discussing the work instead of having to wait for each helper in turn.
Less clutter on screen. The terminal interface now groups its tool activity into collapsible summaries. That makes the answer easier to read; people who want to inspect each step can turn up the detail in settings. The desktop app also adds Jump to message, making it easier to find an earlier request in a long session.
This is also part of the wider OpenCode 2 generation. Anyone moving from OpenCode 1 should check the official migration guide: the command is still opencode, but plugins, server integrations and some configuration rules have changed.
What else is similar to it
Cline is another open-source coding assistant. It has particularly strong editor integrations, including VS Code and JetBrains, as well as desktop and terminal options. It’s worth exploring if you want an AI helper close to the editor you already use.
Aider takes a more terminal-and-Git-centred approach. Git is the tool developers use to track changes to a project. Aider makes those changes easier to inspect and undo, and it can run linting and tests as part of its workflow.
OpenHands Agent Canvas is a different sort of product: a control centre for running and automating coding agents. It can even run OpenCode as one of its backends. It makes more sense when you want to organise several agents or recurring tasks, rather than simply ask for help with one project.
They’re not a ranking from best to worst. A useful first question is: Where do I want to work? In my editor, in a terminal, with a desktop companion, or across several agents? That choice will tell you more than star counts.
Why it’s different
OpenCode’s strength is how much choice it gives you without insisting on one particular AI model or one working style. You can use the desktop app, work directly in the terminal, switch between Plan and Build, or extend it with tools and integrations as your needs grow.
For a newcomer, the nicest use case may be the simplest one: open a practice project and ask, “Can you explain what these files do, in plain English?” For an experienced developer, that same tool can help explore an unfamiliar codebase, prepare a change or coordinate several smaller tasks. It can grow with the user instead of demanding all the setup up front.
There are two boundaries worth understanding. First, review changes: an AI can misunderstand a request, just as a person can. Plan mode is useful for discussing changes, but it is not a full security sandbox. The permissions guide explains how to require approval before commands or access to other folders. Second, consider privacy: the AI model you choose may receive parts of your project to answer you. Don’t begin with passwords, customer data or a sensitive workplace repository.
That is not a reason to avoid OpenCode. It’s simply a good reason to start small, learn what it is doing, and give it more responsibility when you’re comfortable.
How to get started
- Choose your starting point. The desktop app is friendlier if you prefer clicking through windows; the terminal version is there if you already use command-line tools. Download from the official OpenCode website.
- Prepare a small practice folder containing something simple, such as a text README and a tiny JavaScript file. Avoid private documents, passwords and work projects for your first experiment.
- If you chose the terminal version, install it with
npm install -g @opencode/cli(requires Node.js). If you prefer not to use commands, the desktop download skips this step. - You’ll need an AI model to answer your questions. OpenCode offers free options, and also lets you connect other providers. Check whether the model you select has a price before you begin.
- Open OpenCode and select your practice folder. In the terminal, go to that folder and type
opencode. The app should open a conversation connected to those files. - Choose an AI model. In the terminal interface, type
/connectto add a provider if needed and/modelsto choose a model. In the desktop app, follow its model or account controls. - Select Plan mode before your first question. The terminal interface lets you switch modes with
Tab. Try this: “I’m new to this project. Please explain what the files do in everyday language. Don’t change anything.” - Read the explanation and ask follow-up questions. When you’re ready for a first edit, switch to Build and try something tiny: “Please add one sentence to the README explaining what this project is for.” Look at what changed before moving on.
- No answer? Make sure a model is selected. In the terminal, try
/models, then/connectif your provider isn’t connected. Some providers also need an account or payment method. - The terminal says
opencode: command not found? Check that the installation finished and reopen the terminal. Or use the desktop app, which avoids this command-line setup. - OpenCode asks to run something you don’t recognise? You can stop and ask it to explain the command first. A request to “do nothing” is not a technical permission rule; use the permissions settings for stronger boundaries.
- Already installed OpenCode 1? Follow the OpenCode 2 migration instructions before replacing it. Both versions use the
opencodecommand, so it’s worth checking which one you have.
Further reading: Official getting-started guide · Connecting an AI model · Permissions and approvals · What changed in OpenCode 2.3.0 · Source code on GitHub. Information checked 15 November 2026.
