@cline/sdk Turns Cline’s Runtime into a Shared Foundation
Cline has released @cline/sdk, an open-source agent runtime that now powers every Cline interface and is available for any team to build on. Instead of bolting new features onto IDE-specific code, the team rebuilt the core agent loop as a standalone, portable SDK and migrated their own CLI and Kanban products on top of it, with VS Code and JetBrains integrations following. This move effectively decouples the agent brain from any particular editor, making it a reusable runtime rather than a tightly coupled plugin. For developers, it means they can treat Cline not just as a finished AI coding assistant, but as a base layer for open-source coding agents adaptable to their own workflows. The same infrastructure that serves more than 7 million developers is now packaged as a general-purpose agent runtime SDK, installable via npm install @cline/sdk.
Layered Agent Runtime SDK Avoids Vendor Lock-In
Under the hood, @cline/sdk is a layered TypeScript stack, with each package focused on a single responsibility. @cline/shared provides core types and utilities, while @cline/llms manages the provider layer for Anthropic, OpenAI, Google, AWS Bedrock, Mistral, LiteLLM, and other OpenAI-compatible endpoints. Crucially, this provider logic is kept completely outside the agent loop, so switching models becomes a configuration change instead of a code fork. @cline/agents runs the stateless agentic loop, handling iterations, tool orchestration, and event emission, and @cline/core wraps it with stateful orchestration for session lifecycle, persistence, and configuration discovery. App surfaces like the CLI, VS Code, and JetBrains now connect to this runtime instead of owning it. Teams can pull the full stack or only the packages they need, creating custom coding agents without tying themselves to a single proprietary platform.
Durable Sessions and Long-Running Work for AI Development Tools
Rebuilding the runtime as a portable SDK has direct implications for long-running work and serious AI development tools. Sessions are no longer tied to a single interface or IDE window; they can survive UI restarts and move across surfaces such as the CLI and Kanban. The agent loop stays strictly stateless and reusable, while the surrounding runtime becomes durable and portable enough for multi-hour or multi-day tasks. Alongside this architectural shift, the team rewrote prompts, tightened context management, and rethought how tools are presented to the model. Benchmark results suggest these changes matter: Cline CLI running claude-opus-4.7 scores 74.2% on Terminal Bench 2.0, versus 69.4% for Claude Code on the same model. On open-weight models, Cline CLI hits 55.1% with kimi-k2.6, compared to OpenCode at 37.1% on the same benchmark, signaling competitive performance for an open runtime.
Agent Teams, Plugins, and Connectors for Custom Coding Agents
Beyond the core loop, @cline/sdk ships with features that make it easier to build sophisticated, custom coding agents rather than single, monolithic bots. The runtime supports agent teams and subagents natively, letting a session delegate work to specialist agents, track progress, and exchange handoff notes without an external orchestration layer. A plugin system allows teams to inject domain-specific behavior without forking the core: plugins can register tools, listen to lifecycle events, impose rules, and shape the context presented to the model. Features like scheduled CRON jobs, checkpointing, web search, and MCP connectors are supported out of the box. The new CLI also introduces experimental connector channels via cline connect, so agents can surface directly in Telegram, WhatsApp, Slack, and other platforms. Together, these capabilities turn the SDK into a flexible backbone for AI development tools tailored to specific teams and workflows.
From Single Product to Community Agent Infrastructure
Cline positions this SDK release as the infrastructure move that underlies its recent product evolution. Originally known for delivering what it calls the first real agentic coding experience with Claude Sonnet 3.5, Cline has since expanded to a terminal-first CLI 2.0 with headless CI support and Cline Kanban for orchestrating multiple agents across a git repository. With @cline/sdk, all of these become applications built on a shared, open foundation rather than isolated tools. For teams, it lowers the barrier to entry for extending or customizing agent capabilities—whether that’s building a bespoke coding assistant, embedding agents into existing dev tooling, or experimenting with open-weight models. It also signals a shift toward community-driven agent infrastructure: instead of depending on closed enterprise platforms, developers can adopt an open-source coding agents stack, wire in their own providers, and grow an ecosystem around a transparent, configurable agent runtime SDK.