Cline SDK

Product Introduction

1. Definition: Cline SDK is an open-source TypeScript agent runtime and framework for building, customizing, and deploying AI-powered coding agents and CLI tools. It provides the core orchestration logic, state management, and extensibility layer that powers the Cline family of products.
2. Core Value Proposition: It exists to provide developers with a modular, production-ready foundation for creating sophisticated, long-running, and extensible AI agents, eliminating the need to build complex agentic orchestration and state management from scratch.

Main Features

1. Modular Plugin Architecture: The SDK's core is designed around a plugin system that allows deep customization. Plugins can register custom tools, intercept agent lifecycle events, add new commands, and modify context. They can be developed as local .ts/.js modules or packaged as reusable npm modules with a cline.plugins manifest, enabling teams to prototype and share capabilities easily.
2. Native Subagents & Multi-Agent Teams: The runtime natively supports creating subagents and multi-agent teams without requiring a separate orchestration layer. Developers can spawn specialized agents (e.g., a linter agent, a documentation agent) within a main session. Each subagent can have its own model, tools, and system prompt, and they can communicate via handoff notes, enabling complex, collaborative workflows directly within the SDK.
3. Unified Provider Layer & Model Support: Cline SDK abstracts LLM interactions through a dedicated provider layer (@cline/llms). It supports a wide range of providers out-of-the-box, including Anthropic (Claude), OpenAI (GPT), Google (Gemini), AWS Bedrock, Mistral, and any OpenAI-compatible endpoint (e.g., vLLM, Together.ai). Adding a custom provider involves implementing a simple ApiHandler interface, ensuring maximum flexibility in model choice.
4. Stateful Runtime with Checkpointing & Cron Jobs: Unlike stateless agent loops, the SDK provides a durable, stateful runtime. This enables critical features like checkpointing (saving and resuming long-running agent sessions) and scheduled cron jobs (e.g., a nightly agent that audits dependencies or runs security scans), making it suitable for persistent, automated workflows.
5. Integrated MCP (Model Context Protocol) Support: The SDK has built-in support for MCP, a standard for connecting agents to external data sources and tools. This allows developers to easily integrate their agents with databases, filesystems, APIs, and other services through a growing ecosystem of MCP servers, significantly expanding the agent's capabilities and context.

Problems Solved

1. Pain Point: Architectural Lock-in and Lack of Extensibility in Early Agent Products. Many first-generation AI coding tools were built as monolithic applications (like IDE extensions) where the agent logic was tightly coupled to the UI. This made the core runtime difficult to separate, maintain, extend, or embed into other surfaces.
2. Target Audience: Software Engineers and Platform Teams building custom internal AI agents, coding assistants, or automation tools. Product Developers creating commercial AI-powered developer tools or CLI applications. DevOps/ProdEng Teams implementing automated code review, dependency management, or infrastructure-as-code agents.
3. Use Cases: Building a support bot that can read and understand a codebase before answering questions. Creating a multi-agent system where one agent writes code and another automatically reviews and tests it. Developing a long-running cron agent that monitors GitHub issues, generates fixes, and opens Pull Requests. Embedding a powerful, customizable agent runtime into a custom IDE or internal developer portal.

Unique Advantages

1. Differentiation: Unlike many agent frameworks that are either research-oriented (lacking production features) or closed-product-specific, Cline SDK is a battle-tested, product-grade runtime that is also fully open-source and modular. It offers a more integrated and feature-complete experience than low-level libraries while being more extensible and portable than closed-platform agents.
2. Key Innovation: The layered, stateful runtime architecture. By cleanly separating the stateless agent loop, the stateful runtime service, and the provider layer, Cline SDK achieves a unique balance. It allows for lightweight, serverless deployments of the core loop while supporting complex, persistent agent applications with checkpointing, cron, and teams—all within the same coherent framework. This "engine" abstraction is what allows it to power diverse surfaces like CLI, IDE extensions, and web connectors from a single codebase.

Frequently Asked Questions (FAQ)

1. Is Cline SDK the same as the Cline VS Code extension? No, Cline SDK is the underlying open-source agent runtime and framework. The Cline VS Code extension is one application built on top of the Cline SDK. The SDK allows you to build your own agents independent of any specific IDE or UI.
2. How does Cline SDK compare to LangChain or LlamaIndex? Cline SDK operates at a higher level of abstraction. While LangChain/LlamaIndex provide excellent low-level primitives for chains, retrievers, and data connections, Cline SDK provides a full, opinionated runtime for long-running, stateful agents with built-in session management, checkpointing, cron, and a native multi-agent system. It can use LangChain tools via its plugin system.
3. Can I use Cline SDK with local/open-weight LLMs? Yes, absolutely. Through its provider layer, Cline SDK supports any OpenAI-compatible API. This means you can easily connect it to local models served via vLLM, Ollama, or LM Studio by simply pointing the configuration to your local server's endpoint.
4. What is required to create a custom plugin for Cline SDK? A plugin is typically a TypeScript/JavaScript module that exports a function implementing the Plugin interface. This function can receive the runtime context and use it to register tools, listen to events (like onAgentStart), or modify the agent's context. The SDK provides examples and a straightforward API for plugin development.
5. Does Cline SDK support deploying agents as REST APIs or serverless functions? The SDK's layered architecture is designed for this. The core stateless agent loop (@cline/loop) can be embedded in a serverless function for request/response use cases. The full stateful runtime can be run as a long-lived service, and you can build a REST API wrapper around its session management capabilities to control agents programmatically.