Continue review

Continue is an open-source coding assistant that runs as IDE integrations and a command-line interface, rather than a single hosted model or standalone SaaS editor. It connects code editors and the cn CLI to language models, which can be hosted by commercial providers or run locally and on private infrastructure.

In a development stack, Continue sits between the developer's editor or terminal and the model and tool services they configure. Its agent, chat, edit and autocomplete features apply model output to code tasks: answering questions with project context, changing selected sections, suggesting completions and working through multi-step tasks with tools. Developers can assign different models to tasks such as Chat, Edit, Apply, Autocomplete, Embeddings and Reranking.

That modular setup addresses a common integration problem: using AI assistance without committing every task to one model provider. Continue supports Visual Studio Code, the JetBrains IDE family and the command line. Local model runners include Ollama, LM Studio, llama.cpp and vLLM.

The main integration surfaces are IDE plugins, the cn command-line interface, model-provider connections and MCP. Continue works with OpenAI, Anthropic, Google Gemini, Mistral AI, Groq, Amazon Bedrock, Azure OpenAI and OpenRouter, as well as generic OpenAI-compatible API endpoints. Its MCP client connects to external data sources, developer tools and custom prompts; examples named by Continue include PostgreSQL, GitHub, Sentry and Snyk.

In VS Code, Chat opens with Cmd/Ctrl + L and inline Edit with Cmd/Ctrl + I. In JetBrains, Chat opens with Cmd/Ctrl + J. Autocomplete runs inline, while Agent mode uses tools to inspect, plan and edit across multi-step tasks. Model roles let teams route different tasks to different models.

The vendor does not name SDKs or programming languages for building against Continue. The documented integration surfaces are plugins, CLI, model endpoints and MCP; confirm SDK support and configuration details in Continue's docs before planning a custom integration.

The pricing page lists Open Source (Solo) at $0 per developer per month. It is free and Apache 2.0 licensed, and supports bring-your-own API keys or local models. The page does not publish seat or usage limits for this option. Starter is listed at $3 per million tokens, so its cost is tied to token use rather than a fixed seat price; the page does not publish a usage limit.

Team costs $20 per seat per month and includes $10 in credits per seat. Other seat and usage limits are not published. Company pricing is custom. Annual prices are not published, and the page lists no add-ons.

For a project using API keys, budget for the model provider's pricing as well as any Continue plan charges, and check how each provider bills the models you route through Continue. Local inference avoids sending requests to a hosted model provider, but running local hardware or private servers has its own infrastructure costs. The vendor does not publish overage rates or plan caps, so ask how billing and access behave when included credits or any applicable limits are reached.

Continue is open-source software under the Apache 2.0 license. Developers can use bring-your-own API keys or local models, and the vendor describes support for self-hosting and air-gapped operation. Named local runners include Ollama, LM Studio, llama.cpp and vLLM; models can run on local hardware or private enterprise servers.

This gives teams options to keep inference within their own infrastructure instead of routing code to a hosted model provider. Model roles and support for generic OpenAI-compatible endpoints also reduce reliance on one model vendor: teams can change which model handles a task, subject to endpoint compatibility and configuration.

The vendor does not publish a dedicated data-export path. Your source code remains in your repository, but teams should check what Continue stores for configuration, context and usage, and how to export or remove that data. Before adopting it, document model endpoints, MCP connections and configuration so another IDE plugin or CLI can replace those integration points if needed.

Continue's configurable model routing, MCP client and local-inference options can fit teams with different privacy and tooling requirements. Before relying on it across an engineering organization, validate the models and endpoints each task uses, the context sent to them, and the access granted to connected tools. The vendor does not publish seat or usage limits, overage rates or reliability targets in the plan details; ask about caps, support and incident handling before budgeting for production use.

Scaling costs depend on the plan and model setup: Starter is priced per million tokens, Team per seat with $10 in credits per seat, and BYO providers may have their own charges. Teams should monitor token consumption and provider billing separately, and check local inference capacity if they run models themselves.

Reddit discussion is mixed. One commenter recommends Continue for model choice and multi-file refactors, while another wishes its developers would debug the change-diff function. A separate commenter says the UI is not as good as Cursor's. Treat these as individual reports, not a reliability measure; test review and diff workflows against your own repository before rollout.