AutoGLM Web Search

Problem It Solves

When a model needs current web information, a plain prompt may not include a runnable retrieval path: call the AutoGLM Web Search API, obtain a local token, build dynamic signed headers, and then parse snippet results into citations. autoglm-websearch packages these steps into one reusable skill.

How It Works

• Credential retrieval: at startup, the script sends a GET request to http://127.0.0.1:53699/get_token and receives Bearer xxx; if the response lacks the Bearer prefix, it adds one automatically.
• Search request: it then sends a POST request to https://autoglm-api.zhipuai.cn/agentdr/v1/assistant/skills/web-search with a body such as {"queries": [{"query": "..."}]}.
• Dynamic signing: each request generates X-Auth-Appid, X-Auth-TimeStamp, and X-Auth-Sign, with the signature calculated by an MD5 rule so stale values are not reused.
• Result processing: the skill summarizes entries from the snippet field, organizes the answer around the user query, and appends reference sources at the end.

Boundaries

It is suitable for web search, quick summarization, and citation-list tasks, not offline knowledge Q&A, long-form analysis, or login-gated scraping. Before using it, confirm that the local token service is reachable, API credentials are valid, and the query is specific enough; if the response is empty or noisy, verify the cited sources manually.