finch-knowledge
v0.25.1
Published
Index local documents and images in LanceDB, then search them from Finch.
Readme
Knowledge Base
Keep documents in Finch, organize them with tags, and find passages locally or ask the assistant to look them up.
Install
npx @finchtoys/minitools add finch-knowledgeEnable Knowledge Base in Finch’s Toolcase. First use downloads the local engines; no extra API key is needed for indexing or search.
Get started
- Open Knowledge Base from the sidebar. Choose a tag and click Add files, or drag files into the page.
- Use a file’s Set category menu to add or remove tags. A file can have several tags.
- Click the magnifier to search the selected tag, or use the book button in the chat toolbar to draft a question for the assistant.
For example: “Find passages about onboarding in the Product tag and cite the sources.”
The library menu lets you change its location, back it up, or refresh changed sources. Removing a file deletes its library copy and index, not the original.
Supported files
Markdown, plain text, Word, PDFs with a text layer, and images. Each file can be up to 64 MB. Export scanned PDF pages as images before importing.
Supported platforms: Apple Silicon macOS, Linux x64/arm64, and Windows x64/arm64.
Keep in mind
- Search by keywords. Search matches wording, not meaning; use terms that appear in your documents.
- Updates are manual. Refresh files imported by the assistant from a workspace. For files added through the page, re-import the changed file with the same name and tag.
- Indexing and page searches stay local. When the assistant retrieves passages, those passages and source paths enter the conversation and may be sent to your model provider.
知识库
把资料放进 Finch,用标签整理,在本地查找片段,也可以让 AI 根据资料回答问题。
安装
npx @finchtoys/minitools add finch-knowledge在 Finch 工具箱中启用 知识库。首次使用会下载本地运行环境,建索引和检索无需额外 API Key。
开始使用
- 从侧栏打开 知识库,选择标签,点击 添加资料,或直接把文件拖入页面。
- 在资料菜单的 设置分类 中勾选或取消标签;一份资料可以属于多个分类。
- 点击放大镜检索当前分类,或点击对话输入框工具栏的书本按钮,选择并编辑提问模板。
例如:「在『产品』标签里查找用户引导相关的片段,并注明来源。」
资料库菜单可更换存放位置、备份或刷新已改动的资料。移除资料只删除库内副本和索引,不会删除原文件。
支持的文件
Markdown、纯文本、Word、含文本层的 PDF 和图片。单文件最大 64 MB。纯扫描 PDF 请先将页面导出为图片再导入。
支持 Apple Silicon Mac、Linux x64/arm64 和 Windows x64/arm64。
使用须知
- 按关键词检索。 匹配文字而非语义,尽量使用资料中出现的词。
- 修改后需更新。 AI 从工作目录导入的文件可通过刷新更新;页面添加的文件,请在同一标签下重新导入改好的同名文件。
- 索引和页面检索在本地完成。 AI 调用资料时,片段及来源路径会进入对话,可能发送给所用模型的服务商。
