Continue.dev acquired by Cursor: migrate to Cline locally (2026)

#What happened

Cursor (the Anysphere company) acquired the Continue team and discontinued the standalone product. According to Bodega One, the announcement dates to around June 16, 2026, and the latest version, 2.0.0 of the VS Code extension, was released on June 19. The README in the continuedev/continue repository confirms the essentials: the repository is no longer actively maintained and is read-only for everyone.

Repository frozen, license unchanged

The code remains under the Apache 2.0 license (2023 to 2026, Continue Dev, Inc.). You can clone it, fork it, and run the version you already have.

Latest version with no telemetry or authentication

The maintainers state that the final 2.0.0 release removes anonymous telemetry and authentication, making it usable without an account with a local configuration.

Closed hosted service

According to Bodega One, data stored with Continue (history, synced configurations, team settings) was deleted after July 15, 2026. Anything on your disk was unaffected.

JetBrains: CLI recommendation

The README recommends using Continue’s CLI instead of the JetBrains plugin.

#Should you migrate, and when?

Contrary to what some alarmist headlines suggest, nothing stops on your machine overnight. An extension already installed and configured locally with Ollama will continue to work. The question is therefore one of timing and risk, not absolute urgency. Here is how to decide.

The security issue deserves consideration without alarmism. An editor extension reads your files and can execute code; without a maintainer, a discovered vulnerability won't be fixed. For a personal project on an isolated workstation, the risk is low. For enterprise code, it becomes a real governance issue.

#Cline vs Continue: what really changes

Cline’s repository describes it as an open-source coding agent for the editor, terminal, and desktop. Its license is Apache 2.0, not MIT as is sometimes stated; it is also a permissive license with a patent grant, with no practical impact for you. It accepts Ollama and LM Studio, as well as any OpenAI-compatible API.

#And Roo Code, Kilo Code, and the others?

Roo Code, long cited as Cline's most active fork, ceased operations on May 15, 2026, according to Bodega One, and recommended Cline itself for those who want a model-independent open-source extension. Kilo Code, a Roo Code fork, remains a possible alternative. To stay closest to a living lineage with Ollama, Cline is the most direct choice; if you prefer the terminal, OpenCode and Aider do the same job without an editor.

• OpenCode + Ollama: a coding agent in your terminal
• Roo Code: the local coding agent (project discontinued)

#Which coding model to choose

Cline talks to Ollama: any coding model available in Ollama will work. The criterion is memory, including context. The sizes below come from the Ollama library and cover the weights alone, before the context cache. A coding agent sends a lot of text (open files, directory trees, command output): memory headroom matters more than for simple chat.

#Step-by-step migration

1. 01
   Export what still exists

   If you had a Continue account, check whether any data remains to be recovered. If your configuration is a local file, simply copy it: it contains your models and custom prompts, which you can copy into Cline.
2. 02
   Verify that Ollama responds

   Run ollama list. Ollama must listen on http://localhost:11434, the address Cline uses by default.
3. 03
   Download a coding model

   Choose based on the table above, then pull the model with the command ollama pull.
4. 04
   Adjust the context

   Start Ollama with a 64,000-token context, or set it in the Ollama app. Check with ollama ps that it still fits entirely on the card.
5. 05
   Install Cline

   In VS Code, open the Extensions panel, search for Cline, and install it. An icon appears in the sidebar.
6. 06
   Choose Ollama as the provider

   In Cline’s settings, select the Ollama provider, keep the http://localhost:11434 URL, and choose your model from the list.
7. 07
   Enable the compact prompt

   Cline’s documentation recommends enabling Use Compact Prompt (Settings, Features) for local inference.
8. 08
   Uninstall Continue once everything works

   The two extensions can coexist during testing; remove Continue afterward to avoid shortcut conflicts.

ollama pull qwen3.5:9b
OLLAMA_CONTEXT_LENGTH=64000 ollama serve

If Ollama is already running as a service (macOS, Windows), do not use the second command: adjust the context length in the Ollama application settings, or you may run into a port conflict. On Linux with systemd, define the variable in the service.

#What carries over, and what does not

Your model choices and the Ollama address can be copied into Cline’s settings in a minute. Your custom prompts and style guidelines require manual work: rewrite them as project or task instructions, taking the opportunity to shorten them, since a local agent responds better to concise instructions. Keyboard shortcuts change. As for your Continue conversation history, there is no evidence that a import automatique exists: consider it non-transferable and export anything valuable before uninstalling.

#The settings that make Cline work locally

Cline’s documentation for local models provides RAM guidelines: 16 to 32 GB for small quantized models, 32 to 64 GB for medium-sized code models, and more than 64 GB for large models and large contexts. It also recommends keeping tasks focused, because a smaller context makes responses faster, and starting a new task when the context becomes too large.

In practice, a local agent behaves like an intern with a short memory: give it a file and a precise objective, not “refactor the entire project.” If the first response is slow, check with ollama ps that the model isn't split between the GPU and CPU; that's the most common cause of slowness.

#Restore online autocomplete

If grayed-out text while typing matters to you, install a dedicated extension alongside Cline, also connected to Ollama: Tabby or Twinny. They generally use a small model specialized in completion, which doesn't need the agent's large model and therefore saves memory.

• Tabby: self-hosted local code autocompletion
• Twinny: 100% local code autocomplete

#Keep your data local

Check the active provider

In Cline’s settings, the provider must be Ollama or LM Studio. This setting alone determines whether usage is local or sent to a remote server.

Configuration on your drive

No account is required to use Cline with a local model: there is no new hosted service to depend on.

Offline test

Once the models are downloaded, disconnect from the network: if the agent responds, no outbound traffic is required for it to operate.

• Cline + Ollama: 100% local coding agent in VS Code
• Best local LLM for coding
• Source: Continue repository (read-only)
• Source: Cline documentation, local models
• Source: Ollama documentation, context length
• Source: Cline repository

#Frequently asked questions