Verdict. I installed it to see how it compares. I’m staying on Claude Code. The idea I’d take from it is the plan agent, which can’t edit files because its definition denies writes, so the boundary doesn’t depend on the model behaving.
what it is
An open-source terminal AI coding agent in the same category as Claude Code and Aider. MIT licensed, bring your own model.
It’s a TypeScript and Bun monorepo. The agent runs as a local client and server pair, and the TUI, a beta desktop app and the SDKs all talk to the same server.
Two built-in agents, switched with Tab.
- build is the default and has full access.
- plan is read-only. It denies file edits and asks before running bash.
There’s also a general subagent you call with @general for multi-step searches.
Install with curl -fsSL https://opencode.ai/install | bash, or through npm, brew, scoop, choco, pacman, mise or nix.
what I found
| Stars | 200,660 (25,948 forks), GitHub API, 2026-08-23 |
| Contributors | 456 |
| Commits | about 100 in six days (2026-08-17 to 08-23) |
| CI | 25 GitHub Actions workflows, including test, typecheck and automated triage bots |
| Release | v1.18.21 (2026-08-21) |
- The repo moved from
sst/opencodetoanomalyco/opencode. The old path returns a 301 to the new one with the same star count and no fork parent, so it’s the canonical repo after an org transfer. The Homebrew tap is nowanomalyco/tap/opencode. - I found it on an Instagram list of “GitHub repos that shouldn’t be free.” It’s one of the most-starred coding agents there is, which says more about the list than the repo.
- The default branch is
dev, notmain. - The README asks any project using “opencode” in its name to state it isn’t affiliated.
compared to
- Claude Code is what I run every day.
- Codex is my second model for reviews.