Building Long-Running Agents with ScaleDown and Deep Agents

LangChain has introduced a langchain-scaledown integration and LangSmith LLM Gateway support for ScaleDown’s constrained decoding models, which are specialized for context-management tasks rather than open-ended generation. ScaleDown offers Summarize, Compress, Extract, and Classify models that can process up to one million tokens and, for several tasks, images and scanned PDFs; the company reports substantial speed, cost, and accuracy advantages over larger general-purpose models on narrow benchmarks. The integration provides middleware for replacing older agent history with summaries, reducing tool outputs to material relevant to a query, and extracting exact structured fields and classifications from user messages for use by tools. Developers can also use classification in custom middleware to detect agent looping or drift and stop or redirect runs based on confidence thresholds. ScaleDown is available through the LangSmith gateway, allowing centralized authentication, tracing, spending limits, and rate controls, while general-purpose LLMs remain recommended for planning, coding, and other open-ended tasks.

Use this post, company, and trend context to find content marketing opportunities, perform competitive analysis, or address product feature gaps via the Plushcap MCP server or the Plushcap API.