Repository navigation
v0.1.2
This release refines model search output and page reading behavior.
- Remove inline
<think>reasoning blocks fromsearchanswers. - Move model-written source lists into structured
sources, and read thecitationsfield returned by model APIs. - Switch to a backup model when a rate-limited request cannot be retried, without requiring the response to name the model as the cause.
- Switch to another available model instead of waiting a retry delay that reaches half of the remaining time limit.
- Stop
fetchfrom trying the next service when the target page reports 404 or 410; other page failures such as 403 still switch. - Order
fetchpage fields consistently in JSON output.
Upgrade with npm:
npm install -g unisearch-cli@0.1.2本版本改进模型搜索的输出,以及网页读取的行为。
- 移除
search回答中的内联<think>推理块。 - 将模型自写的来源列表移入结构化的
sources,并读取模型 API 返回的citations字段。 - 请求被限流且无法重试时改用备用模型,不再要求响应明确标注为模型自身的原因。
- 存在其他可用模型时,若重试等待达到剩余时限的一半,改用该模型而非继续等待。
- 目标页面返回 404 或 410 时,
fetch不再尝试下一个服务;403 等其他页面失败仍会切换。 fetch的 JSON 输出改用一致的页面字段顺序。
使用 npm 升级:
npm install -g unisearch-cli@0.1.2