Hacker News spots OpenCode, an open-source AI coding agent built for terminal, IDE, and desktop

The Hacker News discussion around OpenCode was one of the largest AI tooling threads in this crawl, reaching 1,238 points and 614 comments as of 2026-03-20T21:03:52.000Z. The linked project site describes OpenCode as an open-source AI coding agent that works in the terminal, inside an IDE, or through a desktop app. Its public code is hosted on GitHub.

The first major selling point is provider flexibility. OpenCode says it can connect to Claude, GPT, Gemini, and other providers, and that it supports 75+ providers through Models.dev, including local models. The site also highlights GitHub Copilot login and ChatGPT Plus/Pro login, which positions the product less as a single-model experience and more as a reusable shell for whichever coding models a developer already pays for or runs locally. In a market crowded with bundled agent products, that interoperability is an important part of the pitch.

The second theme is workflow integration. OpenCode advertises LSP support so the agent can load the right language servers for a project, and it offers multi-session operation so multiple agents can work in parallel on the same codebase. It also supports share links for sessions, which can be useful for debugging, handoff, or review. Those features suggest the team is aiming beyond a simple chat interface and toward a practical agent layer that fits normal engineering loops.

Distribution is another part of the story. OpenCode is available as a terminal interface, an IDE extension, and a desktop beta, with the site naming macOS, Windows, and Linux for the desktop build. That matters because the coding agent category is fragmenting across shells, editors, and standalone apps. Instead of forcing developers into one surface, OpenCode appears to be trying to keep the same agent experience available wherever a team already works.

The project site also makes aggressive growth claims. It says OpenCode has 120K GitHub stars, 800 contributors, 10,000+ commits, and 5M monthly developers. Those are project-site claims rather than independently audited platform metrics, but they show how the team wants the project to be perceived: not as a niche experiment, but as a large and fast-moving open-source coding stack. The same page stresses a privacy-first design and says OpenCode does not store user code or context data, which may matter for enterprise or regulated environments.

That mix of model choice, LSP integration, multi-session workflows, and privacy positioning helps explain why Hacker News reacted strongly. OpenCode is not just promising better completions; it is trying to present a full agent runtime that can sit across terminal, IDE, and desktop use cases. Readers interested in the story can follow the HN thread, browse the official site, and inspect the GitHub repository to see how much of that platform vision is already in place.