Continue vs opencode

Where opencode wins

It is the replacement for the piece that actually died

Continue's extension still runs offline. Its CLI does not, because the endpoint it was written against is unreachable, and no configuration flag brings a switched-off host back. opencode is terminal-first by design and needs no vendor endpoint on the default path, so the automation you had around cn has somewhere to go. It will be a rewrite rather than a repoint, but it is a rewrite onto something that is being released weekly.

Provider breadth, and an explicit statement about your keys

opencode connects 75-plus providers through Models.dev, local models included, and accepts a GitHub Copilot or ChatGPT account as a sign-in. Its docs say that when you use your own OpenAI or Anthropic keys the tokens are billed directly by the provider and not by Zen, while other Zen models stay available next to them. That sentence is the reason this is a credible destination for a Continue user: the bring-your-own-key path is documented, not tolerated.

The paid gateway calls itself optional, in writing

OpenCode Zen is a curated model gateway with per-token rates and some free models, and the documentation states that it is completely optional and that you do not need it to use opencode. Treat the surrounding marketing as marketing: the project's site claims 195,000 stars, 950 contributors and over 16 million developers a month, and only the star count is independently checkable, where the real figure on 2026-08-21 was in fact higher at 199,678. Usage claims of that shape are not verifiable from outside.

MIT, and it says you own the output

The licence is MIT across the repository, and the docs go further than most in this category by stating that you own all code produced by opencode, with no licensing restrictions or ownership claims. That is a small thing until it appears in a procurement questionnaire.

Where Continue still wins

opencode has not rebuilt the editor. Continue was a plugin in VS Code and JetBrains with a sidebar and inline completion, and opencode is a terminal agent. If your team's use of Continue was ghost text and a chat panel, moving to a terminal agent is not a replacement, it is a different job with different ergonomics, and half the team will quietly stop using it. In that case read the Cline comparison instead, which is the editor-side answer; note that Cline now ships a CLI, an SDK and a desktop build too, so the terminal choice between the two is about design rather than availability.

There is also a fair point about structure, and it is the point this whole site keeps returning to. There is a comfortable story going around that opencode and Cline were never at risk because there was no founding team to buy out. It is false. opencode is owned by Anomaly, listed in Y Combinator's 2021 batch, with a paid gateway and per-seat enterprise pricing, and the project has already changed homes twice, from opencode-ai/opencode (archived, last pushed 2025-09-18) to sst/opencode and now to anomalyco/opencode. It has more commercial machinery attached to it than Continue ever did, and no published governance document. What protects you is the MIT licence on the code you have, plus the fact that the default path needs no server, which is precisely the same protection Continue users are living on right now.

Migrating Continue to opencode

The path most teams take:

1. Inventory first. Separate what you had in the editor from what you had in scripts. The editor half needs a different destination; the script half is what moves here. Anything that came from Continue's Hub is gone, because that host has no DNS record and the export window closed 2026-07-15.
2. Rewrite the automation, do not port it. Your cn invocations and any Continue SDK calls assumed api.continue.dev. Rebuild them against opencode's own command surface and configuration, and add your provider keys directly rather than starting with Zen.
3. Run both for one cycle. Keep Continue installed for the editor work while the terminal work moves over. Tokens are billed by your provider in both cases, so the overlap costs usage rather than two subscriptions.
4. What does not move. Inline completion, your Continue chat history, Hub rules and assistants, and any assumption that a vendor endpoint will be there next year. Pin the opencode version you standardise on, and write down which model each job uses, so the next migration is an afternoon rather than a week.