← back to the archiveCover illustration for “OpenCode: an open-source AI coding agent for the terminal”
TERM6mo ago·Published · Updated ·by Andy Padia

OpenCode: an open-source AI coding agent for the terminal

In short: OpenCode is an open-source AI coding agent for the terminal and desktop that works with models from many providers and is not tied to one vendor.

OpenCode is an open-source AI coding agent. Like Claude Code, it reads your project, edits files and runs commands from natural-language tasks, and it is provider-agnostic: its website, read in November 2026, lists 200+ LLM providers, including local models. It runs as a terminal interface, a desktop app or a web app, and its code is on GitHub under the MIT licence.

OpenCode in plain words

It is the same kind of coding helper, except you can open the bonnet, change the engine and see exactly how it is wired.

How it works

OpenCode provides an interface, an agent loop, tools for editing files and running commands, and configuration for which provider and model to use. Its documentation has sections for MCP servers, instructions, skills, commands and permissions, so the patterns you learn in one coding agent largely carry over.

Example

As an illustration of a typical start, using the install command shown on the site in November 2026 (check the docs for current options):

npm install -g @opencode/cli
cd my-project
opencode

Then choose a provider, give it a task such as "add input validation to the signup handler and write tests", and review the changes it proposes.

Where you'll meet it

  • Developers who want one agent across several model providers
  • Teams that need to run local or self-hosted models for privacy
  • People comparing coding agents on cost and output for the same task
  • Open-source contributors extending agent tooling

What it changes for you

Open-source agents make the harness itself inspectable, which matters for security reviews and for learning how agents work. For teams, provider choice becomes a configuration setting, so model cost and quality can be compared on real tasks. My view: the trade-off is support, because you own more of the setup.

OpenCode vs Claude Code

Claude Code is Anthropic's own agent for Claude models. OpenCode is open source and model-agnostic. The fair comparison is the same task, same repository, same model where possible.

Common mistake

My view: comparing agents on different models and crediting the result to the tool. Hold the model constant to judge the harness, or hold the harness constant to judge the model.

Claude Code · Google Antigravity · Agent harness · Agentic coding · Open-weights model

From the archive: QM makes the coding agent a swappable part

Hold the model constant to judge the harness, and hold the harness constant to judge the model.

#field-guide#coding-agents#open-source#developer-tools#opencode

Sources

  1. opencode.ai
  2. opencode.ai/docs
  3. github.com/anomalyco/opencode
← older drop
MCP server: how a tool exposes itself to AI apps
newer drop →
Batch inference: cheaper AI for work that can wait

related drops

explore all 557 drops →
← back to the archiveday 175