From Coding Tool to Open Infrastructure Layer
Cline’s release of @cline/sdk marks a strategic shift from being just a consumer-facing coding assistant to becoming an infrastructure layer for agent-powered development. The open-source agent SDK now underpins every Cline surface, including the CLI and Kanban interface, with VS Code and JetBrains extensions migrating next. Instead of binding agent behavior tightly to a specific IDE, the team rebuilt the core coding agent runtime as a portable TypeScript stack that any product can embed. By exposing the same runtime that drives its own tools, Cline invites teams to build custom coding agents tailored to their workflows, repositories, and deployment environments. This architecture decouples the agent loop from user interfaces, allowing sessions and capabilities to move fluidly across terminals, editors, and external channels. In the process, Cline positions itself less as a single app and more as a common runtime developers can standardize on, avoiding vendor lock-in at the platform level.
A Layered Coding Agent Runtime Built for Portability
At the heart of @cline/sdk is a layered design that turns the coding agent runtime into modular, reusable infrastructure. The @cline/shared package provides types and utilities, while @cline/llms abstracts the provider layer so Anthropic, OpenAI, Google, AWS Bedrock, Mistral, LiteLLM, and any OpenAI-compatible endpoint become configuration choices instead of code branches. Above that, @cline/agents runs the stateless agentic loop, handling iteration, tool orchestration, and event emission. The @cline/core layer adds stateful orchestration, managing session lifecycles, persistence, and configuration discovery. App surfaces like the CLI or IDE extensions connect at this top layer without owning the runtime itself. Teams can install the full open-source agent SDK via npm install @cline/sdk, or selectively pull packages to keep their integration minimal. This separation of concerns makes it easier to embed a robust coding agent runtime into existing tooling while preserving flexibility over providers, tools, and storage.
Durable Sessions, Agent Teams, and Long-Running Workflows
Rebuilding on a standalone SDK fundamentally changes what coding agents can do over time. Sessions no longer die when a UI restarts, and the same long-running task can move between the terminal, Kanban, and future IDE integrations. The stateless agent loop stays consistent, while the surrounding runtime becomes durable and portable across environments. Cline’s coding agent runtime also bakes in advanced orchestration features often reserved for separate platforms. The SDK supports agent teams and subagents natively, letting a session delegate tasks to specialists, track progress, and exchange handoff notes without an external orchestrator. Plugins can inject domain-specific tools, observe lifecycle events, enforce custom rules, and shape what the model sees. Built-in capabilities like scheduled CRON jobs, checkpointing, web search, and MCP connectors give teams the primitives to construct reliable, long-running development workflows on top of the same open-source agent SDK that powers Cline’s own products.
Performance, Connectors, and the Push to Democratize Coding Agents
Beyond architecture, Cline is using @cline/sdk to push competitive performance and broad accessibility for custom coding agents. In Terminal Bench 2.0, Cline’s CLI running claude-opus-4.7 scores 74.2%, compared with 69.4% for Claude Code on the same model. On open-weight models, the Cline CLI reaches 55.1% with kimi-k2.6, outperforming OpenCode’s 37.1% on an equivalent run. These results suggest that the redesigned agent loop, tighter context management, and reworked prompts meaningfully impact real-world coding tasks. On the access side, the new CLI introduces experimental connector channels, enabling agents built on the SDK to surface in Telegram, WhatsApp, Slack, and other platforms through a cline connect wizard. Combined with Cline’s reach to over 7 million developers, this positions the open-source agent SDK as a neutral foundation for agent-powered development. Teams can own their infrastructure, data, and provider choices while still tapping into a shared, evolving agent runtime.