The announcement also says ScaleDown is supported in the LangSmith LLM Gateway, where calls are traced and count toward spend and rate limits. The agent’s primary model remains responsible for planning and task execution; ScaleDown handles selected context operations. LangChain reports that a tool-result trace reduced a documentation page from 18,402 to 3,110 characters. Performance and price comparisons in the post are attributed to ScaleDown and are not independently verified here.
Raw SignalOct 13, 20261 min read1 source
Building Long-Running Agents with ScaleDown and Deep Agents
In one minute
LangChain announced a `langchain-scaledown` integration for LangChain and Deep Agents, adding middleware for summarizing older conversation history, compressing tool results, and extracting structured data.