Description

The built-in vision_analyze tool fails when using the GLM (ZhipuAI) vision provider (glm-4v-flash) with error:

Error code: 400 - {'error': {'code': '1210', 'message': 'API 调用参数有误，请检查文档。'}}

Root Cause

Request format mismatch between Hermes' vision_analyze tool and the GLM API. The GLM vision API accepts image data in OpenAI-compatible format with content as an array of {"type": "text/..."} and {"type": "image_url", "image_url": {"url": "data:image/...;base64,..."}} objects.

Manual curl calls with exactly this format work correctly (tested with glm-4v-flash at https://open.bigmodel.cn/api/paas/v4/).

Config Used

auxiliary:
  vision:
    provider: glm
    model: glm-4v-flash
    base_url: "https://open.bigmodel.cn/api/paas/v4/"
    api_key: "<valid-key>"

Expected vs Actual

• Expected: vision_analyze tool sends a correctly formatted OpenAI-compatible request to GLM, returns image analysis
• Actual: Returns 400 error 1210 immediately

Workaround

Manual curl + base64 encoding works perfectly (documented in references/glm-vision-provider.md). The GLM API key, endpoint, and model are all correct.

Environment

• Hermes Agent latest (2026-05)
• GLM provider: glm-4v-flash
• Platform: CLI and WeChat gateway (same result, across multiple sessions)