Cline Unveils Open-Source SDK: A Modular Agent Runtime Powering Cross-Platform Development

Introduction

Cline pioneered the agentic approach to AI-assisted coding long before it became mainstream. However, building on the cutting edge often incurs structural debt. Over time, the agent loop became tightly coupled with the VS Code extension, making maintenance and environment migration cumbersome. Adding features onto a rigid core proved increasingly difficult. Now, Cline—the open-source AI coding agent used by millions—has shipped a major architectural overhaul: it extracted its internal agent harness into a standalone, open-source TypeScript SDK called @cline/sdk. All Cline products are being rebuilt on top of this new foundation.

Rather than layering more functionality on the existing architecture, the Cline team chose to rebuild the foundation from the ground up.

What the SDK Is and How It Is Structured

The core agent harness has been abstracted into a pluggable agent SDK (@cline/sdk). It now powers Cline across VS Code, JetBrains, and the CLI, and is open-sourced so any team can build on it. A key behavioral benefit: long-running work no longer dies with a UI restart, and sessions can move across surfaces. The agent loop remains stateless and reusable while the runtime around it becomes durable, portable, and product-agnostic.

Layered Architecture

The SDK is a layered TypeScript stack where each layer has a single responsibility and dependencies flow strictly downward:

• @cline/shared – Foundation package carrying types, schemas, tool helpers, hook contracts, and extension registration utilities. No higher-layer dependencies.
• @cline/llms – Provider gateway and model catalogs. Covers Anthropic, OpenAI, Google, AWS Bedrock, Mistral, LiteLLM, and any OpenAI-compatible endpoint (e.g., vLLM, Together, Fireworks). Provider logic stays outside the agent loop, so switching providers is a config change, not a code change.
• @cline/agents – Browser-compatible, stateless agent execution loop handling iteration, tool orchestration, and event emission. Notably, it does not own session storage, built-in file/shell tools, or Node-specific orchestration—enabling embeddability in browser environments.
• @cline/core – Node runtime/orchestration layer responsible for sessions, storage, built-in tools, hub and remote transports, automation and scheduling, telemetry, and plugin/extension loading.

@cline/sdk re-exports everything from @cline/core. For a smaller footprint, each package is installable individually: you can pull in just @cline/llms for an LLM proxy, or just @cline/agents for a stateless loop in a serverless or browser environment, without the full Node runtime stack.

Performance Improvements

According to the Cline team, internal runs show that the new CLI completes the same tasks faster and at a lower token cost compared to the old version.

Rebuilt Agent Harness

With Cline 2.0, the team also rewrote the underlying prompts to improve agent behavior. While specific benchmark numbers aren't detailed, the structural changes aim to deliver more consistent and efficient agent performance.

Conclusion

Cline's extraction of its agent runtime into an open-source SDK marks a significant step toward modular, cross-platform AI development tools. By decoupling the agent loop from specific editors and runtimes, Cline enables developers to build custom agents, integrate with various IDEs, and run agents in serverless or browser environments—all while benefiting from a stateless, portable core.