From IDE Plugin to Portable Agent Runtime
Cline has released @cline/sdk, an open-source agent runtime SDK that now powers every Cline surface and is available for any team to build on. Instead of extending an architecture tightly coupled to a single IDE, the team rebuilt the core agent loop as a standalone, portable runtime. Cline’s own CLI and Kanban tools already run on this foundation, with VS Code extensions and JetBrains plugins now being migrated. For developers, this marks a shift from siloed plugins to a shared agent core that can span terminals, editors, and visual orchestration tools. Sessions can persist beyond a UI restart and even move across different surfaces, enabling long-running work that isn’t tied to a specific interface. By decoupling the runtime from its host, Cline effectively turns its long-running coding agent experience into a reusable building block for broader developer workflows.
A Layered TypeScript Stack for Custom Coding Agents
The @cline/sdk package is built as a layered TypeScript stack, each layer focused on a single responsibility. At the base, @cline/shared provides foundational types and utilities, while @cline/llms owns the provider layer for Anthropic, OpenAI, Google, AWS Bedrock, Mistral, LiteLLM, and other OpenAI-compatible endpoints. Crucially, provider logic is kept entirely out of the agent loop, so switching models becomes a configuration change rather than a code rewrite. Above that, @cline/agents runs the stateless agentic loop—handling iterations, tool orchestration, and event emission—while @cline/core manages stateful concerns such as session lifecycle, persistence, and configuration discovery. App surfaces like the CLI, VS Code extensions, and JetBrains integrations connect at the top of this stack without owning the runtime. Teams can install the full SDK via npm install @cline/sdk or selectively pull individual packages to tailor their open-source coding agents.
Durable Sessions, Cross-Surface Workflows, and Performance Gains
Re-architecting the agent runtime fundamentally changes how long-running coding tasks behave. Sessions no longer die when a UI restarts and can seamlessly move between the CLI, Kanban boards, and soon VS Code and JetBrains. The agent loop itself remains stateless and reusable, while the surrounding runtime becomes durable and portable across tools. Cline also used the new harness to rewrite prompts, tighten context management, and rethink how tools are exposed to the model. These changes show up in benchmarks: Cline CLI running claude-opus-4.7 scores 74.2% on Terminal Bench 2.0, compared to 69.4% for Claude Code on the same model. On open-weight models, Cline CLI reaches 55.1% with kimi-k2.6, ahead of OpenCode at 37.1% on that run. For teams evaluating agent runtime SDK options, the numbers suggest that an open, standardized foundation can compete with proprietary platforms on real-world coding benchmarks.
Built-In Agent Teams, Plugins, and Multi-Channel Connectors
Beyond the core runtime, @cline/sdk ships with features aimed at complex developer workflows. Agent teams and subagents are supported natively, allowing a session to delegate tasks to specialist agents, track progress, and exchange handoff notes without an extra orchestration layer. A plugin system lets teams add domain-specific behavior—registering custom tools, observing lifecycle events, injecting rules, and shaping what the agent can see—without forking the SDK. Scheduled CRON jobs, checkpointing, web search, and MCP connectors are built-in. The latest Cline CLI introduces experimental connector channels, enabling agents to surface directly into Telegram, WhatsApp, Slack, and other platforms via a cline connect setup wizard. For organizations building developer tools or VS Code extensions, this combination of extensibility and multi-channel delivery turns the SDK into a flexible foundation for embedding open-source coding agents wherever developers work.