openJiuwen's WorkSwarm launches a Personal Knowledge Space for office agents
WorkSwarm, the office agent from openJiuwen, has launched a "Personal Knowledge Space" that ingests user-configured sources including local folders, Edge bookmarks, Zhihu columns, Toutiao, Feishu documents, GitHub and AtomGit, with manual or scheduled incremental collection. The system organizes material into directory-structured, source-linked plain-text knowledge using rule-based, hybrid rule-plus-LLM, or agent-based processing, then feeds it to the agent through progressive disclosure. In a vendor-supplied article published by QbitAI, the openJiuwen team says its own 25-question test set spanning seven source types showed a 32% higher task pass rate, 76.2% less total task time and 84.1% lower task-stage token consumption; the article acknowledges preprocessing itself costs tokens, which can be offset by using a cheaper model for organizing.
It targets the time and token overhead of having agents repeatedly re-find and re-organize personal material, making "collect once, reuse many times" a concrete selling point for office agents.