Cline SDK - Build coding agents with a plugin-based open-source runtime

Cline just open-sourced the agent runtime that powers one of the most-used coding agents in the world.

What it is: @cline/sdk is a TypeScript agent runtime with a plugin architecture, designed to be embedded in any surface VS Code extensions, CLIs, desktop apps, Slack bots, or whatever you're building.

The problem with most agent frameworks is that the loop gets glued to the product surface. Cline had this same issue: the agent grew inside the VS Code extension, and eventually the runtime and the IDE wrapper were inseparable. The SDK is the result of pulling those two things apart: a stateless, reusable loop at the core, with a stateful runtime layer on top of it.

What makes it different: most coding agent runtimes aren't actually portable. They're designed to run inside one product. Cline SDK is explicitly built so sessions can move across surfaces, long-running work survives UI restarts, and the agent loop stays stateless regardless of where it's running. The plugin layer lets teams add domain-specific tools and observe lifecycle events without forking the runtime.

Key features:

• Plugin architecture for registering tools, lifecycle hooks, and agent rules

• Native subagents and multi-agent teams with built-in handoff and orchestration

• Checkpointing, scheduled cron jobs, web fetch, and MCP connectors out of the box

• Provider-agnostic: Anthropic, OpenAI, Google, AWS Bedrock, Mistral, LiteLLM, and OpenAI-compatible endpoints

• Example apps included: Slack bot, VS Code extension, desktop app

Benefits:

• Embed a production-grade agent loop without building the harness yourself

• Swap models and providers via config, not code rewrites

• Multi-agent workflows without a separate orchestration layer

• Start small with individual packages, add the stateful runtime only when you need it

Who it's for: TypeScript developers building coding agents, internal dev tooling, or AI-powered CLI applications who want a stable, extensible runtime rather than a prompt wrapper.

The benchmark numbers on Terminal-Bench 2.0 will get attention, but what actually matters here is that Cline is releasing the same runtime they use internally as an open package. That's a different kind of credibility than a framework built specifically for public release.

P.S. I hunt the latest and greatest launches in tech, SaaS and AI, follow to be notified → @rohanrecommends

Plugin-based runtimes live or die on fault isolation. If a buggy tool crashes mid-session, does it kill the whole agent or is each plugin sandboxed? That's were most SDKs get sloppy.

Honestly Kudos to the Cline team. I have been switching many internal agents over to the new SDK and it has been great! 🫡👏🏻🚀

I like that this focuses on building reusable coding agents instead of only adding AI inside an editor.

The stateless/stateful split makes sense, that's exactly where most agent frameworks quietly fall apart. What happens to tool side effects on session resume? Like if the agent wrote to the filesystem mid-task and the UI restarted, does it know where it left off or does someone have to reconcile that manually?

cline shipping an SDK is the right move - lets people build agents on top of your runtime instead of competing. we use a similar agent runtime at moonshift.io for full app ships. happy to compare notes on what plugin contracts actually need (file system, terminal, deploy hooks, eval loops are the big four).