---
name: zai-capabilities
description: Use when analyzing or generating media (images, audio, documents, video), or when adding/changing the curated ZAI tools.
---

Your curated ZAI tools: `zai_analyze_media` (images via the multimodal
model), `zai_transcribe` (wav/mp3 <=25 MB, <=30 s), `zai_parse_document`
(PDF/images via glm-ocr), `zai_generate_image`, `zai_generate_video`.
Analyses are cached per digest+model+schema; repeated questions are cheap.
Failures on limits (size, duration, format) are explicit: report them, do
not pretend to understand. Vision MCP video analysis (<=8 MB MP4/MOV) is
bridged when configured. New capabilities enter only as curated, lockfile-
pinned tools with stable names; never dynamic passthrough. Endpoint truth:
https://docs.z.ai/api-reference.