opencode Is the Open Coding Agent That Quietly Out-Starred Almost Everyone
Here's a number that should stop you: 178,000 GitHub stars. That's opencode, an open-source coding agent that's now one of the most-starred coding tools on the entire platform, and it's trending again today. It runs in your terminal, reads and writes files, runs shell commands, surfaces LSP diagnostics, and crucially it does a Plan-then-Build split so you review what it intends to do before it touches a single file. Maintained by Dax Raad and the SST team under the Anomaly org, MIT-licensed, 7.5 million monthly developers.
The thing that separates it from Claude Code and Copilot isn't the agent loop, everyone has that now. It's that opencode is provider-agnostic to an almost absurd degree: 75-plus providers, Anthropic, OpenAI, Google, GitHub Copilot, and your own local models, all swappable. You're not buying into one lab's vision and one lab's pricing. You point it at whatever model is best or cheapest this week and keep your exact same workflow. The latest release even folds in opencode-managed provider integration and pipes MCP server instructions into the session context.
It's worth being honest about why we're covering a tool that's been quietly huge for a while: we hadn't, and that's the editorial miss. When something crosses 178K stars and ships 800-plus releases, the open-source community has already voted, loudly, and the signal is that a model-neutral, terminal-first coding agent is what a lot of serious developers actually want to live in.
The bigger pattern is that the coding agent is splitting from the coding model. Anthropic wants you in Claude Code, OpenAI wants you in Codex, and both want you locked to their model. opencode is the bet that the harness should be neutral and the model should be a commodity you swap at will. If that bet keeps winning stars at this rate, the moat moves from the model to the workflow, and that's a very different map of who has power.
Link: https://github.com/anomalyco/opencode
← Back to all articles
The thing that separates it from Claude Code and Copilot isn't the agent loop, everyone has that now. It's that opencode is provider-agnostic to an almost absurd degree: 75-plus providers, Anthropic, OpenAI, Google, GitHub Copilot, and your own local models, all swappable. You're not buying into one lab's vision and one lab's pricing. You point it at whatever model is best or cheapest this week and keep your exact same workflow. The latest release even folds in opencode-managed provider integration and pipes MCP server instructions into the session context.
It's worth being honest about why we're covering a tool that's been quietly huge for a while: we hadn't, and that's the editorial miss. When something crosses 178K stars and ships 800-plus releases, the open-source community has already voted, loudly, and the signal is that a model-neutral, terminal-first coding agent is what a lot of serious developers actually want to live in.
The bigger pattern is that the coding agent is splitting from the coding model. Anthropic wants you in Claude Code, OpenAI wants you in Codex, and both want you locked to their model. opencode is the bet that the harness should be neutral and the model should be a commodity you swap at will. If that bet keeps winning stars at this rate, the moat moves from the model to the workflow, and that's a very different map of who has power.
Link: https://github.com/anomalyco/opencode
Comments