Kimi

What This Integration Enables

Most document automation fails at the chunking step. Splitting a long document to fit a small context window severs the cross-references that make documents mean what they mean: the defined term, the exception two sections later, the amendment that overrides the clause you extracted. Kimi's answer is a context window large enough that you stop chunking: the moonshot-v1 family runs to 128K tokens and the kimi-k flagships extend far beyond, paired with a server-side extraction pipeline that turns PDFs, Office documents, and scanned images into model-ready text. This connector gives FlowRunner agents the whole toolkit, plus the unglamorous action that makes it operationally sane: a token estimator that prices the prompt before anything is spent. - Run multi-turn chat with native tool calling, structured json_object and json_schema outputs, and very long inputs - Turn documents into text in one step, OCR included, with the uploaded file cleaned up automatically - Keep extracted files stored for reuse, or audit and remove them deliberately - Count tokens for the exact model you intend to call, without consuming completion tokens - Discover which models the key can reach, from context-sized variants to vision-capable previews Long context makes whole-document reasoning possible; it does not make the conclusions trustworthy on their own. In FlowRunner flows, what the model extracts becomes a proposal, and the [human-in-the-loop](/concepts/human-in-the-loop) step decides what the business commits to.