A reusable open-source AI agentAn AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.Full definition → runtime now exists with durable sessions that move across surfaces, which you can build on instead of writing your own agent loop.
Key takeaways · AI-distilled
Cline says its original agentic loopThe cycle an agent runs in: decide, call a tool, read the result, decide again — repeating until the goal is met or a stop condition fires.Full definition → grew inside the VS Code extension as features were added, making the runtime hard to separate, maintain or reuse, so it extracted the agent harnessThe scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.Full definition → into @cline/sdk rather than adding more patches.
The SDK keeps the low-level agent loop stateless and wraps it in a durable runtime, so long-running work survives a UI restart; individual packages can be used alone, such as the provider layer for an LLM proxy or the stateless loop for serverless.
Plugins can register tools, observe lifecycle events, and add rules and commands; they can start as a local .ts or .js module and later become a package with a cline.plugins manifest.
Provider logic lives in @cline/llms, separate from the agent loop; adding a custom provider means implementing an ApiHandler and registering it with llms.registerHandler.
In Cline-run pass@1 tests on Terminal-Bench 2.0 as of May 8, 2026, Cline scored 55.1% with kimi-k2.6 versus 37.1% for OpenCode and 45.5% for Pi-Code, though glm-5.1 scored lower in Cline (49.4%) than in OpenCode (52.8%).
Terms in this piece · Glossary
AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
agentic loop — The cycle an agent runs in: decide, call a tool, read the result, decide again — repeating until the goal is met or a stop condition fires.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
