OpenCode launches open-source coding agent with `opencode serve` and multi-backend web UI
OpenCode shipped terminal, desktop, and `opencode serve` workflows for an open-source coding agent with LSP support, plugins, and more than 75 providers. Users should look at the multi-backend web sessions, IPC plugins, and sandboxed local setup as the main differentiators.
TL;DR
- OpenCode's launch page positions it as an open-source coding agent that runs in the terminal, IDE, and desktop app, with LSP support, shareable sessions, and access to more than 75 model providers including local backends.
- The workflow detail that stands out in the Hacker News thread is
opencode serve: users describe a web UI that can be reached remotely and can connect to multiple OpenCode backends at once. - Community examples in the discussion roundup show OpenCode being extended through IPC plugins, including a "self-modifying hook system" and tools that prune and later recover compressed conversation history.
- The same thread also points to a local-first ops story: users report running OpenCode in a sandbox and inside terminal setups such as Emacs vterm while reviewing changes in Magit.
What shipped and how are people using it?
OpenCode | The open source AI coding agent
OpenCode is an open source AI coding agent that assists developers in writing code via terminal, IDE, or desktop app. Features include LSP support, multi-session capability, shareable session links, integration with GitHub Copilot and ChatGPT Plus/Pro, support for 75+ LLM providers including local models, and privacy-first design with no code or context storage. Boasts 120,000+ GitHub stars, 800 contributors, 10,000+ commits, and 5M monthly users. Offers free models, Zen for optimized coding models, desktop beta for macOS/Windows/Linux, and easy installation via curl or package managers.
OpenCode shipped with a broad implementation surface rather than a single editor integration. The launch page says it works across terminal, IDE, and desktop, supports LSP, multi-session workflows, and "shareable session links," and can route across 75-plus providers, including local models. It also claims a "privacy-first" setup with no code or context storage, which matters because many competing coding agents still depend on hosted telemetry or proxy layers.
What makes this launch more concrete is how practitioners in the Hacker News thread are already describing the runtime model. One user says opencode serve is "accessible from anywhere," works with Tailscale, and that the WebUI can attach to "multiple OpenCode backends at once" remote workflow comment. That turns the tool from a local CLI into something closer to a personal coding-agent control plane.
Discussion around OpenCode – Open source AI coding agent
Thread discussion highlights: - everlier on remote/web UI workflows: Uses `opencode serve`, accessible from anywhere, works well with TailScale, and the WebUI can connect to multiple OpenCode backends at once. - khimaros on plugins and context management: Uses OpenCode as a primary harness and built a plugin to add a self-modifying hook system over IPC. - planckscnst on plugins and context management: Wrote a plugin with prune/retrieve tools that can compress conversation history into summaries and later recover the original messages.
The extension model is also more than marketing copy. In the discussion highlights, one user says they built "a self-modifying hook system over IPC" IPC plugin comment, while another describes prune/retrieve tools that compress conversation history into summaries and later restore the original messages context plugin comment. A separate practitioner reports running OpenCode "in a sandbox" and inside Emacs vterm for fast review loops with Magit sandbox workflow comment. Together, those examples make the differentiator clear: OpenCode is launching not just as an agent, but as a hackable harness for remote, multi-backend, and locally isolated coding workflows.