Continue.dev Review 2026

What is Continue.dev?

Continue's homepage is now a single notice, "Continue has joined Cursor", and the GitHub repository behind the extension is read-only. The last stable build, 2.0.0, went out on 19 June 2026. Any continue.dev open source AI code assistant review written today has to start there. The code is still public under Apache 2.0, the VS Code Marketplace listing still counts more than 4.38 million installs, and the extension still talks to almost any model you point it at. What has gone is a company shipping fixes. This page is for Indian developers and small teams deciding whether that trade is acceptable.

What Continue still does on your machine after the Cursor deal

Ty Dunn and Nate Sesti founded Continue in San Francisco in 2023, and its Y Combinator profile places it in the Summer 2023 batch, now marked as acquired by Cursor. The product came in three shapes: a VS Code extension, a JetBrains plugin and a terminal client called cn. Inside the editor you get four modes. Chat answers questions about the code you select. Edit rewrites a highlighted block. Agent runs multi-step tasks that read files and call tools. Autocomplete suggests code as you type.

None of this is tied to a single vendor. You list your custom models in a config.yaml file, give each one a role such as chat, edit or autocomplete, and Continue sends each request to the right place. That flexibility is the reason people picked it over closed assistants.

The README describes a final 2.0.0 release for all three clients, which removed anonymous telemetry, pulled out authentication and fixed a batch of bugs. It also recommends the CLI over the JetBrains plugin. GitHub marks 2.0.0 as the latest release; a 2.1.0 build published the same day is tagged pre-release, which is why some store pages show that number. So the JetBrains build exists, but it is the weaker option and nobody upstream will patch it. GitHub currently shows 35.7k stars and 5.3k forks. Because the licence is permissive, any company can fork the code, patch it and ship it internally without asking permission.

Continue dev with Ollama local setup, step by step

Running models on your own laptop is the use case that keeps Continue relevant. The official Ollama guide asks for 8GB of RAM at minimum, 16GB or more for comfortable use, and about 10GB of free disk.

• Install Ollama (Homebrew on macOS, the script on Linux, the installer on Windows) and start it with ollama serve. A curl request to localhost:11434 should reply that it is running.
• Fetch models with ollama pull rather than ollama run, and always use the exact tag. The guide suggests qwen2.5-coder:1.5b for completions and llama3.1:8b for chat and agent work.
• Open ~/.continue/configs/config.yaml and add two entries with provider: ollama and apiBase pointing at port 11434. Give Llama the chat and edit roles and Qwen the autocomplete role. If Agent mode refuses to start, add capabilities: tool_use to the chat entry.
• Match the model field to what ollama list prints, character for character. A wrong tag is the usual cause of a 404 "model not found" error.

The docs give a rough memory budget: about 4GB for a 1.5B to 3B model, 8GB for 7B and 16GB for 13B. On a typical 16GB office laptop, a 7B chat model plus the 1.5B completion model is the practical ceiling. Once the weights are downloaded, this local LLM stack needs no internet at all, and the docs carry a separate guide on running Continue fully offline.

Plugging Claude, OpenAI or Codestral keys into the same file

Hosted models go in the same config. The docs have provider pages for Anthropic and OpenAI, so you add an entry with the provider name, the model ID and your API key, then pay that provider at its own rates. Continue takes no cut.

One warning from the autocomplete docs saves a lot of frustration. Frontier chat models such as GPT-5 and Claude are not trained on the fill-in-the-middle prompt format that tab completion uses, so they make poor suggestions there. The hosted picks the docs name are Mistral's Codestral and Mercury Coder; for local use, Qwen2.5-Coder at 1.5B or 7B. A sensible split is Claude or GPT for chat and agent work, with Codestral or a small Qwen as the autocomplete model.

Why Continue autocomplete goes silent, and the five usual fixes

The documentation points to a short list of causes:

• No model in config.yaml carries the autocomplete role, or that model was never pulled.
• "Enable Tab Autocomplete" is switched off. Click the Continue item in the VS Code status bar to check.
• editor.inlineSuggest.enabled is set to false in VS Code settings.
• GitHub Copilot or another completion extension is active and competing for the same slot. Disable it.
• The model assigned is a chat model that does not understand the completion format.

For anything else, open the VS Code developer console, or read ~/.continue/logs/core.log in a JetBrains IDE. If suggestions only ever cover one line, the docs suggest setting multiline completions to "always" or moving to a larger model.

Continue dev pricing after the Cursor deal

The free part was always the extension, the CLI and the config file. You never paid Continue to use a local model or your own key. Money changed hands only for the hosted layer: Continue Hub for sharing agents and configs across a team, plus managed model credits. The pricing page still loads and still lists three tiers, with no notice about the deal. The homepage, though, carries an FAQ titled "What about my subscription?" with no public answer, and the company itself has joined Cursor. Treat the table below as a legacy listing, not something you can count on buying.

Continue never published a rupee price; the figures above are conversions at roughly ₹94.7 to the dollar. For a solo developer today the real cost is whatever the model provider bills, or nothing beyond RAM and electricity if you stay on Ollama.

Continue dev vs GitHub Copilot for a team that pays for tools

Copilot is the opposite bet: one vendor, managed models and a steady stream of updates. According to GitHub's plans page, Copilot Free gives 2,000 completions and 50 chat requests a month. Pro costs $10 a month (approx. ₹950) and Business $19 per seat (approx. ₹1,800), with admin policies and a no-training commitment on business data.

Continue costs nothing to install, runs any model you choose and can keep every line of code on your own hardware. Copilot brings a support team, a public status page, enterprise controls and new features each month. For a client project where source code legally cannot leave the building, Continue with a local model is still the cleaner answer. For a team that simply wants completions and agents that keep getting better, Copilot is the safer purchase in 2026.

Continue vs Cline: two Apache-licensed agents, only one still moving

Cline is the closest open-source cousin. Its repository also uses Apache 2.0, shows about 67.6k stars, and offers VS Code and CLI clients with a Plan/Act flow that asks for approval before each file edit or terminal command. Cline's pricing page lists the open-source version as free with no seat fees. You bring keys from providers such as OpenAI or Anthropic, or buy inference credits. The pricing page puts its JetBrains extension, SSO and team dashboard in a custom-priced Enterprise tier, although the README lists a JetBrains plugin alongside the other clients.

The deciding difference is maintenance. Cline has a company behind it shipping releases; Continue is frozen. Cline's pages say nothing about inline tab completion, though, and that was Continue's strongest feature. Some developers now run Cline for agent tasks and keep Continue purely as a local completion engine.

Code privacy and offline work on a frozen codebase

With a local model, prompts and code never leave your machine, and the 2.0.0 release stripped out anonymous telemetry. If you add a hosted key, that provider's data policy applies, not Continue's. Apache 2.0 also permits commercial use, so agencies can use it on paid client work.

The new risk is security. An extension that reads your whole workspace and can run shell commands in Agent mode will receive no upstream patches. Pin a known version, review what it can execute, and avoid installing look-alike forks from unknown publishers. The docs also mark several context providers as deprecated, including @Codebase and @Docs, which were replaced by letting Agent mode explore the repository and documentation itself.

Our continue.dev open source AI code assistant review verdict for Indian teams

Continue still earns a place in three situations: developers who want a private, zero-subscription assistant on a decent laptop, agencies bound by NDAs that rule out cloud coding tools, and engineering teams willing to fork and own the code. Students on a tight budget can run a small Qwen model on a gaming laptop and get useful completions for free.

It is the wrong pick for teams that need vendor support, anyone working mainly in IntelliJ or PyCharm, and anyone expecting new features. For an actively maintained open-source agent, try Cline. For a managed assistant, Copilot. If you want an AI-first editor and are happy to pay, Cursor, the company Continue joined.

Your first hour with Continue in VS Code

• Install "Continue" from the VS Code Marketplace or Open VSX, then open the Continue panel from the sidebar.
• Decide on local or hosted. For local, install Ollama and pull qwen2.5-coder:1.5b plus one chat model.
• Edit config.yaml with your entries and roles, save, and confirm the models appear in the panel's model picker.
• Turn off Copilot or any other completion extension, then type in a real file and press Tab to accept a suggestion.
• Try Edit mode on a small function before trusting Agent mode with terminal access.

Treat the setup as yours to maintain. Keep a copy of the config file in your dotfiles, note the extension version that works, and check community forks every few months in case one gathers real momentum.