From Single Tool to Shared Runtime: What @cline/sdk Actually Is
Cline has released @cline/sdk, an open-source coding agent runtime that now powers every Cline surface—its CLI, Kanban interface, and soon its VS Code and JetBrains extensions. Instead of bolting new features onto IDE-specific integrations, the team rebuilt the core agent loop as a standalone, portable AI agent SDK. The new runtime is implemented as a layered TypeScript stack, where each layer has a clear responsibility and can be consumed together or individually via npm install @cline/sdk. This architecture means Cline’s own tools no longer own the agent—they simply connect to it. For teams, that shift is significant: the same runtime that underpins Cline’s products is now openly available as infrastructure, effectively turning Cline from a single coding assistant into a flexible platform for building open source coding agents across different development environments and workflows.
Inside the Layered Architecture Powering Cline’s Coding Agents
The new coding agent runtime is organized into four main layers designed for composability. At the base, @cline/shared provides core types and utilities. Above it, @cline/llms manages all model providers, including Anthropic, OpenAI, Google, AWS Bedrock, Mistral, LiteLLM, and any OpenAI-compatible endpoint, keeping provider logic fully outside the agent loop so switching models becomes a configuration tweak rather than a code change. @cline/agents runs the stateless agent loop itself—handling iterations, tool orchestration, and event streaming—while @cline/core manages stateful concerns like session lifecycle, persistence, and configuration discovery. App “surfaces” such as the CLI, VS Code AI tools, JetBrains integrations, and Kanban dashboards connect only at this top layer. This separation makes the coding agent runtime both portable and reusable, letting teams embed it in existing tools or build entirely new experiences without forking Cline’s codebase.
Why an Open Runtime Matters in a Market of Closed Coding Tools
Most commercial AI coding tools ship as closed, IDE-bound assistants, leaving teams little control over how agents behave, where they run, or which models they use. Cline’s decision to open-source its coding agent runtime directly challenges that model. Because @cline/sdk exposes the same core loop that powers Cline’s own CLI and IDE extensions, teams can build custom open source coding agents tailored to specific repositories, compliance rules, or internal tools. Provider-agnostic design means organizations can standardize on Anthropic, OpenAI, or open-weight models without re-architecting their stack. The runtime’s portability also lowers barriers to entry: instead of wiring a bespoke orchestration system, developers can treat the SDK as a drop-in AI agent SDK that already understands tools, context windows, and long-running sessions. In effect, Cline is turning a polished product into a public, hackable runtime for coding automation.
Long-Running Workflows, Sessions That Survive, and Multi-Surface Agents
Rebuilding the coding agent runtime as a stateless loop wrapped in a stateful core unlocks new capabilities for long-running development workflows. Sessions are no longer tied to a single UI instance, so work does not vanish when a window closes or a terminal restarts. A session can move between surfaces—for example, from the CLI into a Kanban board or a VS Code AI tools integration—while preserving context. Cline reports that the new harness also includes rewritten prompts, tighter context management, and improved tool surfacing, with benchmark gains in the Cline CLI when running both proprietary and open-weight models. For teams, this means the same agent can handle extended tasks like multi-branch refactors or CI-oriented automation without being constrained by one interface, making the coding agent runtime as durable as the underlying project lifecycle.
Agent Teams, Plugins, and Connectors: Building Custom AI Workflows
Beyond single-agent loops, @cline/sdk ships with first-class support for agent teams and subagents. Within one session, the runtime can delegate tasks to specialist agents, track progress, and exchange handoff notes, all without requiring an external orchestrator. A flexible plugin system lets teams add domain-specific behavior: plugins can register tools, hook into lifecycle events, add custom rules, and shape what the model sees. Features like scheduled CRON jobs, checkpointing, web search, and MCP connectors are built in, turning the coding agent runtime into a foundation for complex automation. The latest Cline CLI also introduces experimental connector channels via cline connect, letting agents surface directly to platforms like Telegram, WhatsApp, and Slack. Taken together, these capabilities position @cline/sdk as a general-purpose AI agent SDK that can power everything from IDE helpers to chat-native devops agents, all under an open-source license.