Identity first
智谱清言 (Zhipu Qingyan) is an AI assistant reached through chatglm.cn. The interface and company naming have changed over time; the consumer service should not be confused with the family of open-weight or API models bearing GLM and ChatGLM labels. Keeping a separate product record prevents a model capability from being misreported as a free consumer feature.
Tasks available in the interface
The official interface includes chat, research reports, presentation production, reading assistance, data analysis and a directory of agents. Some entries route to image and video experiences; those specialized generation systems belong to the corresponding modality teams. Their presence in a menu does not establish that an anonymous international user can access them without verification.
Useful first-use exercise
Use a publicly available company report and ask for its organizational structure. Request that the response distinguish figures quoted from the report from inferred advice. Next ask the assistant to list the exact document passages behind three claims. Repeat with a different writing style and check whether facts remain stable. Save the verified evidence outside the assistant before composing a decision brief.
Account governance and data
An official chatglm.cn legal policy describes text, image, audio and document inputs. It says conversation inputs and generated outputs may be logged, that account history is synchronized, and outlines controls to delete conversations and manage optional memory. These statements require direct reading in Chinese; they are not blanket guarantees for a corporate account or another Zhipu service.
Regional data boundaries
The located privacy policy says personal information collected through the identified service within China is stored domestically and describes cross-border processing disclosures if relevant. That cannot be generalized to every GLM API deployment or a reseller. Users abroad should confirm local access, identity requirements and actual data route before uploading proprietary work.
What citations can and cannot show
The interface warns that AI-generated content needs careful verification. This matters most for financial figures, academic citations, legal rules or factual historical claims: a fluent answer may reflect an old page or missing attachment. In a research report, insist on independently resolvable source links, dated statements and an explicit unknowns section.
Export and lock-in
The policy explains chat-history access and deletion but does not establish a universal structured export of all transcripts, files and custom agent configurations. If workflow portability matters, paste accepted findings into an independent document system and confirm any native export individually rather than promising it exists.
Other products to consider
Doubao and Yuanbao offer similarly regionally focused general assistants; Qwen Studio and iFlytek Spark have their own account and model boundaries. Compare by language, data controller, accessibility and source inspection. Exact release day, paid quotas and international availability are not asserted here.
Sources & Further Reading
- Current consumer assistant UI — https://chatglm.cn/
- Official 2024-06-19 use specification and AI limitations — https://chatglm.cn/UsageSpecification?lang=zh
- Official service privacy and retention scope; different name noted — https://chatglm.cn/legal/zhiqiPrivacy