The best Continue alternatives, compared honestly
Continue was a good deal for people who wanted an assistant they controlled: bring your own key, models declared in a config.yaml, VS Code and JetBrains, Apache-2.0. Then a notice appeared on continue.dev saying Continue had been acquired by Cursor. The site's own JavaScript hard-codes 2026-06-15 as the start of the wind-down clock, the final release shipped 2026-06-19, and the data-export window closed 2026-07-15. The code is still on GitHub under the same licence and the local path still runs; everything that phoned home does not. This page separates the two, then compares seven maintained alternatives on the axis this story actually turned on: who could switch them off.
The best Continue alternative depends on which part of Continue you are replacing:
- Hardest to switch off → goose: Apache-2.0 under the Linux Foundation's Agentic AI Foundation since 2026-04-07, the only project here with a GOVERNANCE.md, and no vendor meter.
- Still working inside the editor → Cline: Apache-2.0, bring your own key, and the most active release line in the set (v4.1.11 on 2026-08-21).
- Terminal-first → opencode: MIT, 75+ providers including local models, with its paid Zen gateway documented as optional.
- Staying on Continue → defensible if your config points at your own key or Ollama: nothing will break, and nothing will ever be fixed either. Narrower cases are in the table below: Zed if you want copyleft and will change editor, Tabby if it must all run on your hardware, Aider if you want a tool with no company behind it at all.
Why teams are moving
What actually stopped working
Continue did not get worse at its job. It stopped having a maintainer, and the parts of it that lived on someone else's servers were switched off on a published schedule. The split between the two is clean and checkable, so here it is with the evidence attached, verified 2026-08-21.
The wind-down has a date, and it is in the code
Continue's own site bundle hard-codes new Date("2026-06-15T00:00:00Z") and adds 30 days to render the export deadline of 2026-07-15. Archived snapshots put the homepage flip between 2026-06-13 06:11 UTC and 2026-06-16 00:01 UTC. The last stable release, v2.0.0 for VS Code, plus a v2.1.0 prerelease, went out on 2026-06-19.
The Hub is not down, it is gone
hub.continue.dev and registry.continue.dev have no DNS records at all. That was the marketplace of shared assistants, rules, model blocks and integrations, last archived 2026-01-15. app.continue.dev returns HTTP 402 from a disabled deployment, and continue.dev/export now 404s.
The hosted API took the CLI and SDK with it
api.continue.dev still resolves, to 15.197.206.2, but nothing answers on port 443; the connection simply times out. It is referenced in 18 files in the repository, 15 of them under extensions/cli/ and packages/continue-sdk/, which is exactly what the cn CLI and the SDK were built against, plus the repository's own GitHub Actions workflow at .github/workflows/run-continue-agent.yml. The commercial product being sold in June was source-controlled AI checks on pull requests, not the extension.
Open in name, frozen in practice
The repository was never archived: it still reports archived: false, issues and pull requests are open, and community pull requests were filed as recently as 2026-08-21. But 456 pull requests are open. 218 have been filed since the final release on 2026-06-19, and 179 of those are still sitting open. Three pull requests have been merged since the final release, the last two on 2026-07-21: one moving the docs site to a custom domain, one removing the retired sign-in link, and one stripping the GitHub issue-reporting entry points from the UI. All three are teardown, not development.
JetBrains users are stranded a release back
The last JetBrains build is 1.0.67, dated 2026-03-27. The June cleanup that removed the retired login flow and replaced Hub slugs in the default configuration shipped only to the VS Code 2.x line. If you are on JetBrains you are running a March build that still assumes infrastructure which no longer exists.
No fork picked it up
Of 5,261 forks the largest has 121 stars, and that one is PearAI vendoring Continue as a submodule rather than continuing it; the next two have 21 and 11. Zero releases since 2026-06-19, zero npm publishes since 2026-06-18, zero marketplace updates. Waiting for a community fork is not a plan.
The pattern
The third one in six months, and licences stopped none of it
Continue is not an isolated event. Roo Code (24,331 stars, Apache-2.0) archived its repository, with its last push on 2026-05-15, after a final release of v3.54.0; its last commits were literally about redirecting roocode.com to roomote.dev. The successor from the same company, Roomote, is licensed under the Fair Core License, Version 1.0, ALv2 Future License (FCL-1.0-ALv2), which is source available rather than open source until each release converts to Apache-2.0 on its second anniversary. A 24,000-star Apache-2.0 project was traded for a source-available one. Void (28,834 stars) archived its repository, last pushed 2026-06-02, while voideditor.com still returns 200 and still offers a beta download. Twinny is archived too, last pushed 2025-08-07. github.com/sourcegraph/cody now returns 404.
Every one of those projects had a permissive open-source licence, and every licence did exactly what it says: it kept the source available and did nothing else. Apache-2.0 is irrevocable, so nobody can take Continue's code back. It did not keep the Hub online, it did not keep accounts working, and it did not give a single person permission to merge one of those 456 pull requests. The licence protects the code. Governance decides whether anyone is still working on it.
The honest counter-example belongs here too, because it cuts against the argument. Kilo Code was acquired by Anaconda on 2026-07-15, and Anaconda's own press release says Kilo remains available with no changes to existing products, plans or support. It is MIT, it shipped v7.4.23 on 2026-08-20, and it is still shipping daily. An acquisition is not automatically a shutdown. What it is, reliably, is a decision made by someone other than you.
The shortlist
7 Continue alternatives worth evaluating
Continue did two separable jobs: inline completion and chat inside an editor you already used, and a config file that pointed at any model you liked. Nothing in this list rebuilds both exactly. So the ranking optimises for what the last two months proved matters most, custody and maintenance, and then tells you what form factor you are accepting. Every pick lists one honest strength and one real weakness, and every price carries its billing cycle.
Apache-2.0 with no enterprise carve-out, 53,114 stars, v1.46.0 released 2026-08-12, pushes daily. The repository is no longer block/goose: it moved to aaif-goose/goose under the Linux Foundation's Agentic AI Foundation, formed 2025-12-09, with the move dated 2026-04-07. It is the only project here that ships a GOVERNANCE.md, with contributor, maintainer and core-maintainer roles written down, and it publishes no seat price or model meter of its own. Weakness: it is a general-purpose agent for the terminal and desktop, not a Continue-shaped editor extension, so there is no inline tab completion and no sidebar-chat parity. Different ergonomics, not a drop-in.
Apache-2.0, 66,572 stars, and the most active release line in the set: v4.1.11 landed on 2026-08-21. It is no longer only a VS Code extension either: there is a GA CLI on npm (cli-v3.0.56), an SDK and a desktop build, all tagged the same week. Bring your own key to any provider, or run local models. The paid paths are optional and Cline's docs say so: pay-as-you-go Cline credits, and ClinePass at $9.99 per month standard rate, currently $4.99 for the first month, for more usage on popular open coding models. Weakness: Cline is Cline Bot Inc., one company with one founder-CEO, which announced $32M in combined Series A and seed funding on 2025-07-31 led by Emergence Capital and Pace Capital. There is no GOVERNANCE.md, no CLA and no trademark policy in the repository. That makes it the most conventionally acquirable project in this shortlist, which is a risk worth naming rather than denying.
MIT, and the largest star count in the category by a wide margin at 199,678 verified on 2026-08-21, with v1.18.19 released 2026-08-20. Models come from 75+ providers through Models.dev, local models included, and you can sign in with a GitHub Copilot or ChatGPT account. Its docs are explicit that the OpenCode Zen gateway is completely optional and that when you use your own keys the provider bills you, not Zen. Weakness: it is a terminal agent rather than an editor extension, and there is a commercial engine attached to the MIT core. The owner is Anomaly, which by its own account was previously SST and is listed in Y Combinator's 2021 batch, selling Zen and per-seat enterprise with no published price; 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.
The only copyleft option here, and that is a genuine structural difference: Zed's source is licensed primarily under GPL-3.0-or-later with Apache-2.0 components where marked, so a fork stays open by force in a way an Apache-2.0 project does not. 88,978 stars, v1.16.1 on 2026-08-19, weekly releases, written in Rust. Personal is $0 forever with unlimited use of your own API keys or external agents, Pro is $10 per month, Business is $30 per seat per month. Weakness: it is a whole editor, not a plugin, so migrating means leaving VS Code or JetBrains rather than swapping an extension. Its README also states plainly that Zed is developed by Zed Industries, Inc., a for-profit company, which raised a $32M Series B led by Sequoia announced 2025-08-20.
MIT across the whole repository with no carve-out, 26,958 stars, v7.4.23 on 2026-08-20, and shipping for VS Code, JetBrains and a CLI. Unusually explicit about bring-your-own-key: the pricing page splits the bill into platform plan, AI inference and cloud compute, with inference via free models, local models, your own keys or its gateway. Free and open source for individuals with usage billed separately, Teams at $15 per user per month, Enterprise custom. Weakness: Anaconda acquired Kilo Code on 2026-07-15. The press release promises no changes to existing products, plans or support, and so far the releases back that up, but the owner is new and the ending is not written yet.
The one built to run entirely on your own infrastructure: completion, answer engine, inline chat and context providers, self-hosted. Community is free for up to 5 users, Team is $19 per seat per month for up to 50 users, Enterprise is custom. Weakness, and it is two: Tabby is not Apache-2.0 as a blanket statement. Everything under ee/ carries the Tabby Enterprise License, which allows copying and modification for development and testing but requires a valid licence for the correct number of seats for production use. And the project is drifting: last push 2026-06-30, no stable release since v0.32.0 on 2026-01-25, the vendor's own what's-new still headlined July 2025, and TabbyML, Inc. has moved attention to a new product called Pochi.
Apache-2.0, 48,371 stars, a terminal tool that talks to whatever endpoint you give it and auto-commits to git. There is no account, no gateway, no meter and no paid tier of any kind, so nobody can sell it out from under you. Weakness: that is the same sentence as its risk. There is no company, no funding and no GOVERNANCE.md, and effectively one maintainer. The last commit was 2026-05-22, 466 pull requests are open, the last GitHub release was v0.86.0 in August 2025 and PyPI aider-chat last moved to 0.86.2 on 2026-02-12. Its own homepage still recommends Claude 3.7 Sonnet and shows a 44K star badge against a real 48,371. Continue's failure mode with fewer servers to lose.
Deliberately left out: Roo Code (archived, last pushed 2026-05-15; its successor Roomote is under the Fair Core License, Version 1.0, ALv2 Future License, source available rather than open source), Void (archived, last pushed 2026-06-02, download page still live), Twinny (archived, last pushed 2025-08-07) and Cody (github.com/sourcegraph/cody returns 404). Amazon Q Developer is a valid commercial answer but it is proprietary and tied to an AWS account, which is the opposite of what someone running Continue with their own key was buying.
Side by side
Continue alternatives compared
Licences read from each repository's own LICENSE file, prices from each vendor's own pricing page, release and star counts from the GitHub API on 2026-08-21. The column that decides this migration is who could switch it off: it is the only one that would have predicted June. Continue's own row is included as the baseline.
| Tool | Form factor | Who could switch it off | Exact licence | Vendor's own paid tier | Latest release | Best for |
|---|---|---|---|---|---|---|
| Continue | VS Code + JetBrains extension | Cursor (Anysphere), which acquired it. Repository not archived, 456 PRs open, 179 of them filed since the final release, 3 merged, nobody merging | Apache-2.0 | None: billing disabled, unused credits refunded | v2.0.0 stable for VS Code, 2026-06-19, plus a v2.1.0 prerelease | A frozen local setup you never intend to change |
| goose | Terminal + desktop agent | Nobody in particular: Agentic AI Foundation at the Linux Foundation, formed 2025-12-09, goose moved 2026-04-07. Has a GOVERNANCE.md | ✓ Apache-2.0, no carve-out | None published | v1.46.0, 2026-08-12 | Not wanting to do this again |
| Cline | VS Code extension, plus CLI, SDK and desktop builds | Cline Bot Inc., one founder-CEO, $32M Series A and seed announced 2025-07-31 | ✓ Apache-2.0 | Credits pay-as-you-go; ClinePass $9.99 per month; Enterprise custom, no published price | v4.1.11, 2026-08-21 | Staying inside the editor |
| opencode | Terminal agent | Anomaly, by its own account formerly SST, listed in Y Combinator's 2021 batch. Already moved orgs twice: opencode-ai/opencode to sst/opencode to anomalyco/opencode |
✓ MIT | Zen gateway, per-token with free models; Enterprise per-seat, no published price; team workspaces free in beta | v1.18.19, 2026-08-20 | Living in the terminal |
| Zed | Standalone editor | Zed Industries, Inc., "a for-profit company" per its README; $32M Series B, Sequoia, 2025-08-20. Copyleft forces a fork to stay open | ✓ GPL-3.0-or-later, with Apache-2.0 components | Personal $0; Pro $10 per month; Business $30 per seat per month | v1.16.1, 2026-08-19 | Wanting the licence to do the work |
| Kilo Code | VS Code + JetBrains + CLI | Anaconda, Inc. It already happened, 2026-07-15; products unchanged so far | ✓ MIT, whole repo | Teams $15 per user per month; Enterprise custom; optional gateway | v7.4.23, 2026-08-20 | Both IDE families at once |
| Tabby | Self-hosted server + IDE clients | TabbyML, Inc., which owns the ee/ licence and has shifted focus to a new product, Pochi. Repo quiet since 2026-06-30 |
✗ Apache-2.0 except ee/, which is the Tabby Enterprise License: production use needs paid seats |
Community free to 5 users; Team $19 per seat per month to 50 users; Enterprise custom | Last stable release v0.32.0, 2026-01-25; only prereleases since | Air-gapped or on-premises |
| Aider | Terminal tool | Nobody: no company, no funding, no CLA. Also nobody maintaining it; last commit 2026-05-22, 466 PRs open | ✓ Apache-2.0 | None at all | v0.86.0 in 2025; PyPI 0.86.2, 2026-02-12 | Scripted, minimal, no vendor |
Star counts on 2026-08-21: opencode 199,678, Zed 88,978, Cline 66,572, goose 53,114, Aider 48,371, Tabby 33,831, Kilo Code 26,958, Continue 35,568. Continue's VS Code Marketplace figure of ~3.96 million is cumulative installs, not active users. Cursor's blog carries acquisition posts for Supermaven, Graphite and Firetiger, but its blog index and sitemap carry no Continue post. Trade coverage reports the deal terms as undisclosed; neither company published them. Prices and free tiers change often; check each vendor for current terms. Compiled August 2026.
Official pages: goose · Cline pricing · opencode · Zed pricing · Kilo Code · Tabby pricing · Aider · Continue on GitHub · Continue docs
A fair call
When staying on Continue is still defensible
Migrating costs a week of muscle memory and a rebuild of every rule and prompt you had shared through the Hub, because there is no export to import from any more. And the surviving local path was not an accident: the final four days of commits removed the dead login requirement and the Login Required interface, replaced Hub slugs in the default configuration with explicit model definitions, dropped the in-app issue-reporting entrypoints, and finally moved the docs to free GitHub Pages. Someone deliberately cut the cord on the way out.
Continue is still fine if…
- Your
config.yamlalready names explicit models and points at your own API key or a local Ollama. It is a client, the licence is irrevocable, and the last release was hardened for exactly this. - You are on VS Code 2.1.0 rather than the frozen JetBrains 1.0.67 build.
- You work offline or under compliance rules where a tool that never changes counts as a feature. The guide "How to Run Continue Without Internet" is still published at docs.continue.dev, along with the config.yaml reference, the Ollama guide and "How to Self-Host a Model".
- You are mid-quarter and would rather migrate deliberately than in a panic. Nothing else is going to break on a schedule; the servers are already off.
Move now if…
- You relied on the Hub for shared assistants, rules or model blocks: that host has no DNS record and nothing replaces it. Rebuild on Cline or goose.
- You are on JetBrains: the plugin stopped at 1.0.67 on 2026-03-27 and never got the de-vendoring. Kilo Code ships both IDE families.
- You scripted the
cnCLI or the Continue SDK: those callapi.continue.dev, which no longer answers. opencode is the terminal-native replacement. - You need a security patch, a new model provider, or any fix at all. 456 open pull requests and nobody with a merge button is the whole answer.
- Custody is now a procurement question for you. goose sits under the Linux Foundation; everything else here is owned by a company.
Common questions
Common questions about Continue alternatives
Is Continue.dev dead?
The company is gone and the product is unmaintained, but the code has not been deleted. A notice went up on continue.dev saying Continue had been acquired by Cursor; the site's own JavaScript hard-codes 2026-06-15 as the start of the wind-down clock. The last stable release, v2.0.0 for VS Code, plus a v2.1.0 prerelease, shipped 2026-06-19, and the hosted service was wound down with a 30-day data-export window that closed 2026-07-15. The extension is still on the VS Code Marketplace and Open VSX, still Apache-2.0, and still works with your own API key or a local model. Nothing has been merged into the repository since 2026-07-21.
Was the Continue GitHub repo archived or made read-only?
No, and this is where a lot of secondary coverage is wrong. As of 2026-08-21 github.com/continuedev/continue reports archived: false, issues and pull requests are open, and community pull requests are still being filed daily. What changed is who can merge. Counted on 2026-08-21: 456 pull requests are open, 218 have been filed since the final release on 2026-06-19, and 179 of those are still sitting open. Three pull requests have been merged since the final release, the last two on 2026-07-21: one moving the docs site to a custom domain, one removing the retired sign-in link, and one stripping the GitHub issue-reporting entry points from the UI. All three are teardown, not development. The repository is open in name and frozen in practice.
What exactly stopped working when Continue shut down?
Everything that called Continue's own servers. hub.continue.dev, the marketplace of shared assistants, rules, model blocks and integrations, and registry.continue.dev no longer have DNS records at all. app.continue.dev returns HTTP 402 from a disabled deployment. The hosted API at api.continue.dev still resolves, to 15.197.206.2, but nothing answers on port 443; the connection simply times out. It is referenced in 18 files in the repository, 15 of them under extensions/cli/ and packages/continue-sdk/, which is what the cn CLI and the Continue SDK were built against, plus the repository's own GitHub Actions workflow. The commercial product, source-controlled AI checks on pull requests, is gone, and continue.dev/export now returns 404.
Can I still use Continue today with my own API key?
Yes, and it was deliberately made to work that way. In the days before the final release the maintainer removed the dead login requirement and the Login Required interface, replaced the Hub references in the default configuration with explicit model definitions, and stripped the in-app sign-in links. The documentation is still live on GitHub Pages at docs.continue.dev, including the config.yaml reference, the Ollama guide, How to Self-Host a Model and How to Run Continue Without Internet. A local config.yaml plus your own key or Ollama needs no vendor at all. What you will not get is security patches, new model providers, or any fix, ever.
Does the Continue JetBrains plugin still work?
It still installs, but it is a release behind the shutdown and that is a real warning rather than trivia. The last JetBrains build is 1.0.67 from 2026-03-27. The cleanup that removed the retired login flow and replaced the Hub-based default configuration shipped only to the VS Code 2.x builds in June, so the JetBrains plugin is a March build that still assumes infrastructure which no longer exists. Expect to define models by hand in config.yaml and to run into sign-in paths that lead nowhere.
Which Continue alternative is least likely to disappear the same way?
goose is the only project in this comparison that is not owned by a single company. It moved from block/goose to aaif-goose/goose under the Linux Foundation's Agentic AI Foundation, which was formed on 2025-12-09, with the move dated 2026-04-07. It is Apache-2.0, it ships the only GOVERNANCE.md in this shortlist, and it has no model meter of its own. The trade-off is real: goose is a general-purpose agent for the terminal and desktop, not an inline-completion extension inside your editor, so it is not a drop-in replacement for what Continue did in the sidebar.
Is anyone maintaining a community fork of Continue?
Not meaningfully. Of 5,261 forks the largest has 121 stars, and it is PearAI vendoring Continue as a submodule rather than continuing it; the next two have 21 and 11 stars. No fork publishes releases. Nothing has been published to npm since 2026-06-18 or to the VS Code Marketplace and Open VSX since 2026-06-19, and no commit has landed on main since 2026-07-21. If you need a tool that gets patched, migrate now instead of waiting for a fork to appear.