Org-Agent: Beyond Personal Assistants Towards Organizational Agents
Org-Agent schedules multi-user organizational tasks using constraint-aware dependency graphs and memory tools.
Org-Agent is a constraint-centric framework for language-model agents that coordinate requests from multiple users while using knowledge spread across their interactions. It decomposes a task into atomic subtasks, builds a dependency graph, schedules subtasks by topological sorting, and executes them under rules for identity, authority, access permissions, attribution, temporal validity, and conflicting requirements. Experiments on MUSES-Bench and GroupMemBench show gains on cross-user decision-making and memory use, and ablations support dependency modeling and tool use.
- Targets cross-user decisions and shared organizational memory
- Constraints cover identity, authority, permissions, and temporal validity
- Decomposes tasks into a dependency graph scheduled by topological sort
- Uses evidence-acquisition and memory-management tools during execution
- Evaluated on MUSES-Bench and GroupMemBench with ablations
From the original article
1 min read at huggingface.co
Language model agents serving organizations must coordinate requests from multiple users while using knowledge distributed across their interactions. We identify two complementary capabilities for this setting, namely cross-user interaction and decision-making, as well as cross-user memory and knowledge use.
By Luyao Zhuang, Yujing Zhang, Zijin Hong · huggingface.co
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.