Continue Review 2026: The Open-Source Coding Agent That Was Acquired by Cursor and Is Now Frozen, Free and Read-Only
Continue was one of the most popular open-source AI coding assistants for VS Code and JetBrains: 4.27 million VS Code installs and 36,099 GitHub stars. Today continue.dev says "Continue has joined Cursor", the GitHub repository is read-only and "no longer actively maintained", and the final 2.0.0 release removed telemetry and sign-in. The code is still Apache 2.0 and still works with local models. This review checks what is true as of Oct 3, 2026, computes the real marketplace rating from raw votes, and tells you plainly when Continue is still a sensible free tool and when it is not.
| Ownership | Acquired by Cursor (page undated; HN thread Jun 15, 2026) |
| Free tier | Everything is free (Apache 2.0); no paid plan to buy |
| Cheapest paid plan | None: you pay your model provider, or $0 with local models |
| Maintenance | Read-only repo; last commit Jul 21, 2026 |
| VS Code rating | 3.25/5 over 185 ratings (API, not the 3.5 star icon) |
Continue is a frozen but working free tool, not a product you can build a team's future on.
"If you are reading an old comparison that says Continue is the open-source Copilot alternative to adopt, stop. The company joined Cursor, the repository is read-only, and there will be no more fixes from upstream. What is left is Apache-licensed code that still runs, especially with local models. That is useful to some people and a risk to everyone else."
My honest read: if you already run Continue with Ollama or your own API key and it works, you can keep using it, and because the licence is Apache 2.0 a community fork is possible. If you are choosing a coding agent today, pick one that is still shipping. The reviews tell the same story from the user side: 50 of the 80 VS Code Marketplace ratings left in the past 12 months are one-star. This is a VERIFIED EDITORIAL REVIEW: continue.dev, its GitHub repository and API, docs.continue.dev, the VS Code, Open VSX, JetBrains and npm listings, and Hacker News threads were checked on Oct 3, 2026. I did not run a controlled hands-on test.
An IDE extension and a CLI with chat, plan and agent modes, plus bring-your-own-model flexibility.
Continue's docs describe it as an "open source AI code assistant for VS Code and JetBrains" that lets developers "create, share, and use custom AI code agents". The pieces: Chat, Plan (read-only tools) and Agent (all tools) modes in a side panel; Edit to change code in place; Autocomplete for inline suggestions; the Continue CLI (cn) for terminal use; and a config.yaml where you pick models per role (chat, edit, apply, autocomplete, embed, rerank) and attach MCP servers and rules. The big difference from most rivals is that you choose the model and where it runs: Anthropic, OpenAI, Gemini, Ollama, Bedrock, Azure, xAI and others, per the docs.
Ownership: continue.dev now reads "Continue (acquired by Cursor)". Cursor itself announced on Aug 14, 2026 that it is now part of SpaceX (Cursor blog, "Joining SpaceX"), so the chain is Continue, then Cursor, then SpaceX. I found no statement that Cursor will maintain or relaunch Continue.
Large installed base, now with no active maintainer.
VS Code Marketplace gallery API "install" statistic, Oct 3, 2026. Cumulative installs, not active users.
JetBrains Marketplace download counter. Cumulative downloads; I deliberately did not use any JetBrains rating field.
Open VSX, version 2.1.0. Cumulative downloads for VS Code-compatible editors.
GitHub, Apache-2.0, 5,444 forks, Oct 3, 2026. Stars show interest, not current usage.
| Date | Event | Source |
|---|---|---|
| 2023-05-24 | continuedev/continue repository created | GitHub |
| 2026-03-27 | Last JetBrains plugin release on the listing (1.0.67) | JetBrains Marketplace; GitHub releases |
| 2026-06-15 | Hacker News thread "Continue has been acquired by Cursor" | news.ycombinator.com/item?id=48548758 |
| 2026-06-19 | Final VS Code releases 2.0.0 and 2.1.0 published; telemetry and sign-in removed per README | GitHub releases; README |
| 2026-07-21 | Last commit on the main branch: "docs: remove Sign in link (login flow retired)" | GitHub commits API |
| 2026-08-14 | Cursor says it is now part of SpaceX | Cursor blog |
| 2026-08-22 | Hacker News: "Continue coding agent is dead. Alternatives?" | news.ycombinator.com/item?id=49398366 |
What I weigh most is that the numbers stop at June. The extension was heavily installed, but the signals that matter for buying today (releases, commits, a maintainer to answer issues) all stop between June and July 2026.
Fans love the freedom to use any model. Recent reviewers mostly report an agent mode that does not finish tasks.
Praise (5 stars, Oct 3, 2026): "Highly configurable and reliable... no restrictions on using your own AI models". Complaints: agent file edits "repeatedly failed with missing or empty filepath arguments" (1 star, Sep 28), "stops following instructions whenever you ask it to do something useful" (1 star, Sep 8), "does not work in remote ssh mode" (1 star, Jul 10), and tab completions "always junk" with local models (1 star, Aug 5).
One commenter had "just settled on using Continue in VSCode to migrate from Copilot Chat not 30 minutes ago". Another woke to an email that Continue had been acquired and said they had leftover credits. A third argued this is "what happens when you do not pay for your tools" (an opinion, not a fact about Continue's finances).
The poster liked it "on vscode with offline Ollama". A reply noted that "'not maintained' is not the same as 'cannot be used'" and that you can fork it. A tiny thread (4 points), but it shows what a local-model user does next: look for a maintained alternative or fork.
There is nothing to subscribe to. The cost is your model provider, or your own hardware.
Continue itself
Apache 2.0 extension and CLI, read-only repo.
- VS Code extension (Marketplace, Open VSX)
- JetBrains plugin (CLI recommended instead)
- Continue CLI (cn)
- No hosted plan to buy
Local models
Ollama or another local server, even offline.
- Docs offer an offline guide
- Autocomplete with small models such as qwen2.5-coder:1.5b
- Agent quality depends on the model
- No data leaves the machine
Provider API
Anthropic, OpenAI, Gemini, Bedrock and more.
- You pay the provider's token rates
- One reviewer said it "rips through tokens"
- No Continue markup to compare against
Hub and paid plans
Login flow retired; pricing page gone.
- README: authentication pulled out in 2.0.0
- Jul 21 commit: login flow retired
- Data deleted per continue.dev
| Item | Price | What you should know |
|---|---|---|
| Continue extension and CLI | $0 | Apache 2.0, per the GitHub README and API |
| Local model via Ollama | $0 plus your hardware and power | Docs say it works with no internet after the models are pulled |
| Hosted model with your API key | Provider token rates | No bundled usage; no Continue subscription needed |
| Continue Hub / paid plans | Not purchasable | Pricing page 404; login retired; docs partly still reference sign-in |
Arithmetic you can check: a day of agent use on a hosted model costs whatever tokens the model burns at the provider's published rates; Continue adds nothing on top. Because the repo is read-only, there is no roadmap that could change that.
Decide by whether you need maintenance or just a free, working local tool.
Keep going, pin your VS Code extension version, and have a fallback plan.
Look at an actively shipping open-source agent instead (see Alternatives).
Continue still fits that narrow case, with small autocomplete models and a capable chat model.
Not Continue. There is no vendor roadmap, no hosted plan and no maintainer for issues.
Continue's owner now makes Cursor, so read the Cursor review. Nothing I read says Continue users get a migration deal.
Apache 2.0 allows it, but you inherit the bugs, the stale docs and the JetBrains and CLI gaps described below.
Chat, Plan and Agent modes, with permission prompts before tools run.
Continue's docs define three modes. Chat has no tools. Plan gives read-only tools for exploring and planning. Agent has all tools for making changes. By default "Agent mode will ask permission when it wants to use a tool", and you can set tool policies to exclude tools or make some automatic. Tool output is fed back into the model as context. If a model does not support native tool calling, the docs say Continue falls back to system-message tools, and the Agent and Plan buttons show "Not Supported" when the model cannot do tools. That last point explains many bad reviews: Agent mode quality depends heavily on the model you pair it with, and small local models often struggle.
| Mode | Tools available | Use it for | Risk |
|---|---|---|---|
| Chat | None | Questions and explanations | Lowest; no file changes |
| Plan | Read-only | Exploring a repo and drafting a plan | Low; reads files it can see |
| Agent | All tools | Edits, commands, multi-step changes | Highest; needs permission policy review |
| Edit / Apply (model roles) | Targeted edits | Refactors inside a file | Review each diff |
| Autocomplete | Inline model role | Tab suggestions | Quality varies with the small local model |
- You pick the model per role, including local ones, and you can run offline.
- Permission prompts and tool policies keep a human in the loop.
- Open code under Apache 2.0 means you can inspect or fork it.
- Agent mode reliability is the most common recent complaint (empty file paths, looping plans, files replaced).
- Docs are partly stale: they still recommend models like Claude Sonnet 4.6, GPT-5 and Gemini 2.5 Pro and show older model names.
- No upstream fixes are coming for any of it.
Pick a model, plan first, approve each tool call, review the result.
Because upstream will not change, the workflow is a fixed one. Pin the extension version you tested, keep your config.yaml in version control, and treat Agent mode as something to watch rather than leave running. For a more guided path, the Claude Code setup guide shows what a maintained terminal agent workflow looks like.
Continue's own install walkthrough, the clearest primary-source view of the extension UI that the sections above describe.
Published by the official "Continue" YouTube channel (https://www.youtube.com/@continuedev). Vendor video and likely dated: it predates the 2.0.0 final release, so UI details may differ.
The cn CLI runs the same agent in a terminal or in scripts, with its own login confusion.
The Continue CLI (cn) is described as "a terminal-based coding agent" that can edit files, run commands and work through multi-step tasks, "the same agent that powers the Continue IDE extensions". It has two modes: an interactive TUI and headless mode (cn -p "prompt") for scripts, CI and git hooks. Flags include --auto (allow all tools without prompting), --readonly (plan mode), --allow and --exclude for specific tools, --rule, --mcp and --model. The npm package @continuedev/cli shows version 1.5.47 as latest (published Jun 18, 2026).
| Job | How | What to watch |
|---|---|---|
| Interactive terminal agent | cn in your project | Docs ask you to log in with Continue or enter an Anthropic API key; the README says authentication was removed, so check what your installed build asks for |
| One-shot script or CI step | cn -p "prompt" | Headless runs can use --auto; that means no permission prompts |
| Read-only exploration | --readonly | Safest flag for unfamiliar repos |
| Resume a session | --resume | Resumes the most recent session |
| IDE alternative | JetBrains plugin | README recommends the CLI instead; listing shows 1.0.67 from Mar 27, 2026 as latest |
- Scriptable, local-model friendly automation for people who like terminals.
- Plan-only runs for safe exploration.
- Docs and README disagree about login, so setup instructions may not match the build.
- Headless
--autoremoves the permission step. - No maintainer to confirm or fix a CLI bug.
MCP servers and any model provider are the integration story.
Continue supports Model Context Protocol servers so the agent can reach external tools and databases, configured as mcpServers in config.yaml. Model providers include Anthropic, OpenAI, Gemini, Ollama, Amazon Bedrock, Azure and xAI. The docs page for its own MCP server and an older hub that listed MCP blocks both exist, but the hub's login flow is retired, so treat hub-hosted blocks as unavailable until you confirm. If MCP is new to you, JAVIS's Claude MCP and Connectors guide explains the idea.
| Integration | What it does | Status as read |
|---|---|---|
| MCP servers | External tools and databases for Agent mode | Documented in docs.continue.dev; configured in config.yaml |
| Model providers | Anthropic, OpenAI, Gemini, Ollama, Bedrock, Azure, xAI and more | Documented; you supply keys; docs model lists are partly dated |
| Ollama / local servers | Offline coding assistant | Dedicated guide plus a "run without internet" guide (telemetry toggle described) |
| Hub blocks and assistants | Shared configs and MCP blocks | Login flow retired; treat as unavailable |
| Open VSX | VS Code-compatible editors | Version 2.1.0, 1,659,681 downloads |
The only money in play is token cost or hardware.
| What you are doing | What you pay / hit |
|---|---|
| Chat and agent on a local model | $0 in fees; you pay with hardware, power and slower or weaker output |
| Agent on a frontier model through your key | Provider token rates; one recent reviewer said it "absolutely rips through tokens" (anecdote, not a measured rate) |
| Autocomplete | Docs suggest small local models such as qwen2.5-coder:1.5b for speed |
| A team rolling it out | $0 licence, but no vendor support, no hosted admin, no SSO |
| Switching away later | Your config.yaml and MCP setup are portable in concept, but expect to rebuild them by hand in another tool |
Telemetry is gone and the code is open, but nobody is patching it.
| Control | What the sources say |
|---|---|
| Licence | Apache 2.0 (GitHub and README) |
| Telemetry | README: final 2.0.0 removed anonymous telemetry; the offline guide still describes switching off PostHog telemetry, so the docs lag the code |
| Data held by Continue | continue.dev: "All user data has been deleted in accordance with the privacy notice, except... legal, regulatory, tax, accounting, security, or other legitimate business purposes" |
| Agent permissions | Asks before using a tool by default; tool policies; CLI has --readonly, --allow, --exclude and --auto |
| Local models | Can run fully offline per the docs; no code leaves the machine |
| Maintenance | Repo is read-only; last commit Jul 21, 2026; no upstream security fixes should be assumed |
The practical read: the privacy story is better than most (no telemetry in the final build, local models possible), and the maintenance story is worse than most (none). For a personal laptop with local models that is acceptable. For work code that touches production credentials, prefer a tool with a security contact. This is not legal or security advice.
Best for solo developers who want a free, local-model assistant and accept a frozen tool.
| Situation | Why Continue fits (or does not) | Likely plan |
|---|---|---|
| Hobbyist with Ollama on a good GPU | Free, offline, any model; Agent quality varies with the model | Free (local) |
| Developer who wants autocomplete with a small local model | Docs recommend small coder models; user reports are mixed | Free (local) |
| Existing user with a working setup | No need to move today; pin versions and plan an exit | Free |
| Someone choosing a first AI coding tool | Pick a maintained tool; Continue has no future updates | See Alternatives |
| Team needing support, SSO or compliance | No vendor, no hosted plan; Cursor or Copilot fit better | Not Continue |
| Developers on JetBrains | Plugin stalled at 1.0.67 (Mar 2026) per the listing; README prefers the CLI | Use the CLI or a maintained plugin |
Six real gaps, not manufactured ones.
The GitHub README says the repository is "no longer actively maintained and is read-only for all users". The last commit on the main branch was Jul 21, 2026. Bugs reported today will not be fixed upstream.
Recent Marketplace reviews describe failed file creation, looping plans and replaced files. 50 of 80 reviews in the last 12 months are one-star (computed from the reviews API).
The CLI quickstart still asks you to log in with Continue; the README says authentication was removed. The offline guide still discusses telemetry, and model recommendations name older models.
The README says a final 2.0.0 covered the VS Code extension, CLI and JetBrains plugin, but the JetBrains listing shows 1.0.67 as latest and npm lists the CLI at 1.5.47. I could not reconcile that.
The pricing page is a 404, login is retired and user data was deleted. Anything that depended on the hub or an account needs a local replacement.
Continue joined Cursor, which says it is now part of SpaceX. None of that gives Continue users a roadmap.
Also noted by reviewers: setup takes longer than rivals, remote SSH sessions do not work well, and one person said it has no OpenRouter support (I did not test either claim).
If you want maintenance, pick a tool that still ships.
| If you mostly need... | Compare Continue with... |
|---|---|
| An AI-first editor from Continue's new owner | Cursor review |
| An open-source agent in your editor, still shipping | Cline review, Kilo Code review |
| A fast editor with agents built in | Zed review |
| An assistant bundled with your Git host | GitHub Copilot review |
| A terminal agent and a workflow guide | Claude Code setup guide |
| MCP basics before wiring tools in | Claude MCP and Connectors guide |
| Entry point | Continue | Zed (JAVIS review) | Cline (JAVIS review) |
|---|---|---|---|
| Free option | Free, Apache 2.0, read-only repo | Free Personal plan | Open source |
| Maintained today | No | Yes per its review | Yes per its review |
| Paid tier | None | Pro $10 and Business $30 per its review | See its review for ClinePass, Teams, Enterprise |
| Unit that runs out | Your provider tokens | See its review | See its review |
No affiliate relationship shapes this review.
JAVIS has not established a publisher affiliate program with Continue, and Continue sells nothing today. The CTA points to its official page. If anything changes, the review's facts, limitations and alternatives will not.
Go to the official page.
Open continue.devRead the alternative that matches your real question.
Choose the next articleContinue sits among open-source agents that are now choosing new homes.
Every card below links to a published JAVIS review.
Card images come from each product's own site or its JAVIS review page.
Questions people are actually searching right now
Is Continue still maintained?
No. The GitHub README says the repository "is no longer actively maintained and is read-only for all users". The last main-branch commit I found is Jul 21, 2026, and the final releases (VS Code 2.0.0 and 2.1.0) shipped Jun 19, 2026.
Was Continue acquired by Cursor?
Yes. continue.dev says "Continue was acquired by Cursor". The notice is undated; Hacker News discussed it on Jun 15, 2026. Cursor then announced on Aug 14, 2026 that it is now part of SpaceX.
Is Continue free?
Yes. The software is Apache 2.0 and there is no paid plan to buy. You pay your model provider if you use hosted models, or nothing in fees if you use local models.
Can I still install and use it?
Yes. The VS Code extension is on the Marketplace and Open VSX (version 2.1.0, updated Jun 19, 2026) and the CLI is on npm. It just will not receive upstream fixes.
What happened to my Continue account and data?
continue.dev says all user data has been deleted in accordance with its privacy notice, except where retention is required for legal, regulatory, tax, accounting, security or other legitimate business purposes. The login flow was retired.
What is the real VS Code rating?
The Marketplace API reports 3.25 out of 5 across 185 ratings, which the page shows as about 3.5 stars after rounding. Reviews from the last 12 months average 2.20 across 80 reviews, computed from the raw reviews API.
Does it work with local models?
Yes. The docs include Ollama and "run without internet" guides. Reviewers report that agent mode and autocomplete quality depend heavily on the model and are often disappointing with small ones.
What are the best Continue alternatives?
Cline or Kilo Code for open-source agents still shipping, Zed or Cursor as editors, and GitHub Copilot for a vendor-supported assistant. See Alternatives above.
Where this came from, and when it was checked.
Status and ownership: continue.dev (FAQ answers included), github.com/continuedev/continue README, commits and releases API (Oct 3, 2026), Cursor blog "Joining SpaceX" (Aug 14, 2026).
Product: docs.continue.dev (overview, agent quick start, CLI quickstart, models, MCP, install, Ollama and offline guides); npm registry for @continuedev/cli; JetBrains Marketplace (downloads and update dates only).
Market: VS Code Marketplace gallery API (4,270,100 installs; averagerating 3.254 over 185 ratings), Open VSX (1,659,681 downloads), GitHub (36,099 stars, 5,444 forks, Apache-2.0).
Real users: the VS Code Marketplace reviews read in full pages (173 of 185 retrievable; computed average 3.14; last 12 months 80 reviews, average 2.20, 50 one-star); the Marketplace page's own rounded star display was not used; Hacker News items 48548758, 48580147 and 49398366. The JetBrains rating field was not used. Trustpilot and G2 were not checked; Reddit was not available to us.
Alternatives: live JAVIS reviews of Cursor, Cline, Kilo Code, Zed, GitHub Copilot, Claude Code and Claude MCP.
Official video: youtube.com/watch?v=gjekTru4Cg4; author "Continue" on YouTube.
JAVIS Fit Score (38/100): editorial composite of free price and local-model freedom (strong), open licence and privacy posture (good), agent reliability per recent user reviews (weak), maintenance and roadmap (very weak: read-only), documentation consistency (weak), ownership and support (weak). Weighted by judgement; no automated test was run.
