Cline
The open-source AI coding agent that runs in your editor — on your models, under your control.
By Shaun · Co-founder · ToolNav
Quick Verdict
Cline is the best open-source AI coding agent — an Apache 2.0, model-agnostic VS Code extension where you bring your own model and the agent acts only with your approval. It is the most transparent, control-friendly option in the category. The trade-off is cost: you pay for model inference directly (no bundled subscription), so spending scales with usage and you manage it yourself. Ideal for developers who want control and no vendor lock-in; less ideal if you want one flat predictable price.
Cline is an open-source AI coding agent — described by its maintainers as "the open source coding agent in your IDE and terminal." It runs primarily as a VS Code extension (it began life under the name Claude Dev before the 2.0 rename to Cline), and is also available as a CLI and an SDK. Unlike the AI-native editors (Cursor, Windsurf) that fork VS Code, Cline installs into the editor you already use. Its defining traits are openness and control: the code is Apache 2.0 licensed and public on GitHub, it is model-agnostic (you connect Anthropic, OpenAI, Google, OpenRouter, AWS Bedrock, or local models via Ollama/LM Studio), and every action the agent takes — file edits shown as reviewable diffs, terminal commands, browser actions — requires your explicit approval by default. The trade-off for that transparency is that Cline has no bundled all-in subscription: you pay for AI model inference directly (bring your own key, or add usage-based credits through a Cline account), so cost scales with the model you pick and how much you run it.
Our Rating
8/10
Pricing
Free, open-source extension (Apache 2.0) · you pay only for AI model inference — bring your own API key or add credits via your Cline account · cost depends entirely on model and usage
Best For
Developers who want an open-source, model-agnostic coding agent inside their existing VS Code, with full control over which models they use and what the agent is allowed to do
Category
AI Tools
Key Facts
| Rating | 8.0 / 10 |
| Category | AI Coding Tool — open-source agent (VS Code extension) |
| License | Apache 2.0 — open source (github.com/cline/cline) |
| Pricing | Free extension · pay only for AI model inference (BYO key or Cline credits) |
| Models | Model-agnostic — Anthropic, OpenAI, Google, OpenRouter, Bedrock; local via Ollama/LM Studio |
| Key modes | Plan / Act · reviewable diffs · checkpoints · MCP support · per-action approval |
| Best For | Developers wanting an open-source, controllable, model-agnostic agent in their existing VS Code |
Pros & Cons
What works
- Fully open source (Apache 2.0) — the agent runtime is public on GitHub (github.com/cline/cline), inspectable and forkable, unlike proprietary Cursor or GitHub Copilot
- Model-agnostic — connect Anthropic, OpenAI, Google, OpenRouter, AWS Bedrock, or run local models via Ollama and LM Studio; no lock-in to one vendor
- Runs in your existing VS Code as an extension — no editor switch, unlike the VS Code forks
- Plan/Act modes: Plan explores and strategises without touching files; Act executes — with full planning context carried over
- Human-in-the-loop by default — every file edit (shown as a reviewable diff), command, and action needs your approval; checkpoints let you roll back
- MCP (Model Context Protocol) support — connect databases, APIs, and custom tools to the agent
What doesn't
- No bundled subscription — you pay for model inference directly, so cost is usage-dependent and can run high with frontier models on large contexts
- Cost is unpredictable by design — there is no official "typical" monthly figure because it is pure pass-through inference; budget-conscious users must monitor usage
- Requires oversight — the default approval-per-action model is safe but not hands-off (auto-approve exists, but carries the usual risk of autonomous file/command execution)
- Setup friction vs. bundled tools — you provision an API key or Cline credits before you can start, where Cursor/Copilot bundle the model into one subscription
- Quality depends on the model you choose — Cline is the harness, not the model; results track whichever provider/model you point it at
Pricing
| Plan | Price | Best For |
|---|---|---|
| Extension | Free | The Cline VS Code extension, CLI, and SDK — open source under Apache 2.0 |
| Model inference (BYO key) | Pay-as-you-go | Connect your own Anthropic/OpenAI/Google/OpenRouter key — you pay the provider directly for usage |
| Cline account credits | Usage-based | Add credits in your Cline dashboard for unified billing across supported models (app.cline.bot) |
| Enterprise | Custom | Teams needing centralised billing and team features — contact Cline |
Who It's For
Open-source agent, reviewable diffs, explicit approval per action, and full visibility into what the agent does — the most transparent option in the category
Model-agnostic and BYO-key — use your existing model contracts or local models, with no dependency on one vendor's bundled subscription
Installs as an extension in your current VS Code — no migration to a forked editor like Cursor or Windsurf
Cline needs an API key or credits set up first and cost management; Cursor or Copilot are lower-friction if you want one flat subscription
How It Compares
| Dimension | Cline | Cursor | GitHub Copilot |
|---|---|---|---|
| Type | Open-source coding agent | AI-native editor | IDE plugin |
| Licensing | Apache 2.0 (open source) | Proprietary | Proprietary |
| Pricing model | Free + pay-per-inference (BYO key) | Subscription ($20/mo Pro) with bundled models | Subscription ($10/mo Pro) + AI Credits |
| Models | Model-agnostic (Claude, GPT, Gemini, local) | Claude, GPT, Gemini (curated) | GPT, Claude, Gemini (plan-dependent) |
| Control | Approval per action + diffs + checkpoints | Composer agent with diff review | Agent mode + cloud agent |
| Editor | VS Code extension (your editor) | VS Code fork (replaces editor) | Plugin for many IDEs |
Our Rating
Our Verdict
Cline is the clearest answer to a specific question: "I want an AI coding agent, but I want it open-source, on my choice of model, and under my control." It is the most transparent tool in the category — Apache 2.0, model-agnostic, approval-per-action, with reviewable diffs and checkpoints — and it runs in the VS Code you already use rather than asking you to switch editors. The cost of that openness is that Cline is a harness, not a bundle: you bring the model and pay for inference directly, so spending is usage-dependent and you own the cost management that Cursor or Copilot abstract away behind a flat subscription. For developers and teams who value control, want to avoid vendor lock-in, or already have model API access, that trade is a feature, not a bug. For someone who just wants the lowest-friction "install and go" experience at a predictable monthly price, a bundled tool will feel easier. Cline earns its place as the open-source pick in the category on substance, not novelty.
Get ClineFrequently Asked Questions
Is Cline free?
The Cline extension is free and open source (Apache 2.0). What you pay for is AI model inference — you either bring your own API key from a provider like Anthropic, OpenAI, or Google and pay them directly, or add usage-based credits through a Cline account for unified billing. There is no subscription for the extension itself; cost depends entirely on which model you use and how much you run it.
How does Cline compare to Cursor?
Cursor is a proprietary AI-native editor (a VS Code fork) with models bundled into a $20/month subscription. Cline is an open-source agent (Apache 2.0) that installs into your existing VS Code and is model-agnostic — you bring your own model and pay for inference directly. Cursor is lower-friction and more predictable in cost; Cline offers more control, transparency, and no vendor lock-in. The right choice depends on whether you value a bundled experience or open, model-flexible control.
What AI models does Cline work with?
Cline is model-agnostic. It works with Anthropic (Claude), OpenAI (GPT), Google (Gemini), OpenRouter (which exposes many models), AWS Bedrock, and other providers — as well as local models run through Ollama or LM Studio. Because Cline is the agent harness rather than the model, output quality tracks whichever model you connect.
Is it safe to let Cline edit my code and run commands?
By default, Cline requires your explicit approval for every action — file edits appear as diffs you can review, modify, or revert, and terminal commands are shown before they run. Changes are tracked with checkpoints so you can roll back. There is an auto-approve option to let it run more autonomously, but enabling it carries the usual risk of an agent editing files or executing commands without a per-step review, so use it deliberately.
Ready to try Cline?
Get Cline