OpenCode is shipping like a team that wants to own the whole coding workflow, not just the prompt box. ToolVitals records 1,267 GitHub commits in 30 days, 28 release events in 30 days, 30 releases in 90 days, 28 active contributors, 153,674 stars, and a 97 ToolVitals score with a 100 shipping score. That is not the profile of a sleepy open source project.

The product positioning matches the pace. The homepage calls OpenCode the open source AI coding agent, then immediately stretches that into terminal, IDE, and desktop use. It says the tool supports LSPs, multi-session work, share links, GitHub Copilot login, ChatGPT Plus/Pro login, 75+ model providers, and any editor. It also says a desktop app is in beta on macOS, Windows, and Linux. This is a broad bet, not a single-interface toy.

The recent releases show where the team is spending time. v1.14.27 added a configurable default shell for terminals and agent shell commands, trimmed terminal noise while creating TUI workspaces, and hid provider connection checks until onboarding is done. v1.14.26 fixed permission rule ordering, preserved DeepSeek reasoning output, sent a versioned User-Agent header, and improved editor context with Zed support. v1.14.24 added experimental HTTP API endpoints for MCP server status, file listing, file reads, and project file status. The pattern is clear, OpenCode is tightening the boring plumbing that makes an agent usable in real repos.

What the data does not tell you

ToolVitals sees shipping, not taste. It cannot tell you whether OpenCode is reliable under load, whether teams keep using it after the demo, whether the desktop app is better than the terminal, or whether the permissions model actually feels safe. The website also claims 6.5M monthly developers and 850 contributors, but that is a marketing claim, not a measurement ToolVitals can verify. The safe inference is narrower: the team is iterating hard and widening the surface area.

How it compares

Against other developer tools in ToolVitals, OpenCode is moving fast, but not alone. Its 28 release events in 30 days trail OpenClaw’s 48 and PostHog’s 66, but beat LiteLLM’s 24 and Hermes Agent’s 6. On stars, OpenCode sits at 153,674, well behind OpenClaw’s 367,632, but far above Hermes Agent’s 129,983. The signal is not that OpenCode is the biggest project in the room, it is that it is shipping at a pace that keeps it in the conversation.

Bottom line

If your team wants an open source coding agent that has to work in terminals, editors, and desktops, OpenCode deserves a trial. If you only want a thin chat wrapper, skip it. The thing this project is polishing is the stuff that matters in production, shells, permissions, LSPs, sessions, onboarding, and file access.

Sources