← News

Agents & MCPSources: arXiv · 19 Nov 2026

KnowledgeClaw paper proposes omnimodal knowledge engineering benchmark and agentic system

An arXiv abstract introduces OmniKE, a formalization and benchmark for knowledge bases that are omnimodal, closed-loop and multi-granular, plus KnowledgeClaw, an agentic system built on those ideas.

In plain words

Knowledge engineering means turning raw files into a knowledge base that AI systems can reason over. The authors say no existing system or benchmark combines three traits at once: handling all input types, updating itself continuously, and serving answers at different levels of detail. They released a benchmark and an agent-based system intended to do all three.

What was announced

The paper, posted to arXiv by the KnowledgeClaw team, argues that live knowledge bases need omnimodality, a closed-loop lifecycle and multi-granular application simultaneously, while existing systems meet at most two of these and no benchmark measures all three. To address this, the authors formalize Omnimodal Knowledge Engineering and release OmniKE-Bench, which contains six tasks with 1,829 model-free items drawn from 3,960 real files spanning five modalities.

Alongside the benchmark, the authors propose KnowledgeClaw, an agentic system that builds an omnimodal typed hypergraph, keeps it current through a closed-loop lifecycle agent, and routes each query to the granularity it needs. In the reported evaluation against 30 baselines, KnowledgeClaw ranks first on all six tasks with an overall OKE score of 55.9, which the authors state is 14.6 points above the strongest baseline. The software and data are publicly available in a GitHub repository, according to the abstract.

What it does not show

  • The evidence is limited to the paper's abstract; the full methodology, evaluation details and reproducibility of the reported results are not covered by the supplied passages.
  • The performance comparison comes from the authors' own evaluation and has not been independently confirmed.

For security and IT teams

Access
The authors state that the KnowledgeClaw software and data are publicly available on GitHub.

Sources

Source-support review completed 20 Nov 2026. This story was drafted with AI assistance and checked against retained source passages. Automated checking is not independent verification.

Coverage: arXiv: abstract.

Updates and corrections

No corrections so far.

First published 20 Nov 2026

New to the subject? AI from the Ground Up starts from the beginning.