The ZClaw AIOS autumn release is out. Compared with the spring version, the most visible shift is the product focus: from how many models we can connect to how accurate the knowledge actually is.
Change one: the knowledge base now has governance tools
In the past, teams simply uploaded documents and stopped there, which produced uneven retrieval results. The new version splits the knowledge base into four stages: import, cleaning, review and publish. Duplicate and expired documents are detected at import, business tags are applied during cleaning, multi-level approval runs at review, and content only becomes visible to the AI after publishing.
This workflow reflects problems we hit in real projects: without review, the model happily quotes a internal draft to a customer as if it were an official position.
Change two: model routing by scenario
Previously an administrator configured models by hand. Now the system recommends by scenario: support Q&A defaults to a low-cost small model, complex document analysis switches to a stronger reasoning model, and code generation uses a dedicated code model. Switching can follow configured rules or adapt automatically based on observed results.
The most direct benefit is cost: on the same project, scenario-based routing saved more than 40% of token spend compared with using a large model for everything.
Change three: finer support operations reporting
The support module now reports at three levels. The conversation layer shows containment and escalation rates; the knowledge layer shows which questions are answered poorly and what knowledge is missing; the model layer shows which model underperforms in which scenario. These questions used to require digging through logs; now the report suggests the next action.
We ran it internally for a month and found that the suggested knowledge gaps column surfaced more issues than our own review meetings did.
Change four: easier integration with legacy systems
The autumn release adds a set of ready-to-use connectors with preset interfaces for WeCom, DingTalk and common ERP systems. Legacy systems do not need major changes: connectors sync data into the knowledge base and bring AI capability into existing workflows. This answers one of the most frequent customer requests.
How to try it
The autumn release is already available in the demo environment. If you want to see the knowledge governance workflow and scenario-based model routing in practice, request a demo account and bring your own business documents. It is far more convincing than screenshots.