Min Anki	Max Anki	Updated
23.10	26.09+	2026-10-31

GitHub: https://github.com/jrz0213/FastCardMaker

English

FastCard Maker User Guide

Based on FastWordQuery (FastWQ), with the original author credits and license retained.

FastCard Maker fills Anki note fields using AI, local MDX dictionaries or online dictionaries. Use them separately, or choose different sources for different fields of the same note. AI can help with concepts, questions and study material as well as vocabulary.

Start with a few notes before running a larger batch. Screenshots use example fields, dictionaries and models; your names may differ.

1. Create cards with AI

Connect an AI service

Open FastCM → Options → AI settings:

Click Add profile and choose a Provider preset. Presets are available for OpenAI (ChatGPT), Claude, Gemini, DeepSeek and Qwen. For another compatible service, choose Custom.
Give the profile a recognizable name.
Enter the API address, API key and Model. The address must include the endpoint path, such as https://api.deepseek.com/responses. Preset URLs are editable; use a model supported by your service.
Optionally enter shared instructions in Skill, such as “Explain clearly and include a concrete example.” Click OK to save.

Use More settings → Custom parameters (JSON) for reasoning settings. DeepSeek and Qwen both default to Responses, with reasoning.effort set to none in new profiles to disable thinking. Model support for disabling thinking and available levels varies.

These connections also serve dictionary script analysis. Consider a separate “Dictionary parsing” profile with higher reasoning effort for complex scripts. The table contains reasoning settings only; replace the custom parameters with a supported example. Provider names link to official documentation.

Service and APIHigher-reasoning JSON exampleOpenAI (Responses){"reasoning":{"effort":"high"}}DeepSeek (Responses){"reasoning":{"effort":"high"}}Qwen / Model Studio (Responses){"reasoning":{"effort":"high"}}Claude (Messages, adaptive-thinking models){"thinking":{"type":"adaptive"},"output_config":{"effort":"high"},"max_tokens":16000}Gemini (Interactions){"generation_config":{"thinking_level":"high"}}

Use examples supported by your model; high does not imply the same reasoning budget across services. The Claude example retains the required max_tokens output limit, which can be adjusted. Older models may need different thinking settings. Higher effort can increase waiting time and token use; click ? for parameter help.

Region or workspace settings may require a different URL, especially with Model Studio; use the endpoint from your own console.

Map fields and write prompts

Prepare your notes in Anki first. For a concept example, create Topic, Question, Answer and Comparison fields. Enter “Statistics” in Topic and “How do correlation and causation differ?” in Question, leaving the other two for AI.

In FastCM → Options, select the corresponding note type:

Use the radio button on the left to select Topic as the source field.
Keep Topic and Question ignored so their contents are preserved.
Uncheck Ignore for Answer, choose the saved AI connection as its content source, and enter a prompt.
Configure Comparison the same way if needed, with its own prompt. Click OK to save the mapping.

Prompts can reference several fields:

Using {{Topic}}, answer this question: {{Question}}
Start with a direct answer, then give one example. Keep it under 150 words.

Placeholders are replaced with values from the current note. Field names must match exactly, and referenced fields must not also be AI output fields. Put field references in the field’s prompt rather than Skill.

For Comparison, try Compare definitions, examples and common misconceptions about {{Question}} in {{Topic}}, using a Markdown table. Switching the field’s AI connection preserves its prompt.

Vocabulary works the same way: use a word as the source, then request meanings, examples or synonyms in other fields. The add-on fills existing notes. If results do not appear on cards, add the output fields to the relevant Anki card template under Cards….

Test a few notes, then run a batch

Select a few notes in the Anki Browser and choose FastCM → Query Selected. Inspect the results before processing more notes.

For a larger batch, open FastCM → Query whole note types, check the required types, choose a saved configuration for each, then click Query.

Choose note types and saved configurations

This processes all notes of the checked types, regardless of the current deck or Browser selection. Opening from the Browser preselects types represented by selected notes; opening from the main window selects none.

Skip existing values controls whether a populated field is queried again. To regenerate one field, clear that option and ignore other targets. Save changes before running the query.

Sending, formatting and progress

Open Plugin settings → AI settings:

One request per card is the default: fields of the same note using the same AI connection share one request, and results are assigned to their respective fields.
One request per field requests each target separately.
Render Markdown converts headings, lists and tables to card HTML; enabled by default.
Enable CSS injection is off by default. Turn it on and select the note types that should use the shared style. View / edit shared CSS opens the editor. Styles are added after a query actually writes a note of that type; saving settings alone does not modify card styling.

The shared request rate, timeouts and retries are on this page too. Dictionary threads, matching and cloze format are under Dictionary settings; language, scheduling and backups are under General settings. Click ? beside a group for details.

Click Run in background to continue using Anki. Reopen FastCM → Task progress to inspect status and results; Finish task hides the completed progress strip. Results are written about every 30 seconds, with remaining results written when the task finishes.

Closing Anki interrupts the task. If a recovery prompt appears on the next launch, restore replies already received and cached. Unfinished requests are not recovered this way.

Automatically process new notes

Each note type initially uses its first configuration as the starred configuration. To change it, click another configuration’s star in Options and save.

Under Plugin settings → General settings, enable scheduled generation, set the interval, select note types and save. Only empty fields of notes created after enabling the feature are processed. If a batch is still running, the next tick is skipped.

When eligible notes are found, a ten-second countdown offers Run now or Skip this run. Closing the prompt also skips; leaving it alone starts the batch. Keep Anki open and check Task progress for results.

Save settings, back up and use another computer

Click OK to save changes. Closing an edited settings window asks whether to save; an unchanged window closes immediately.

Under Plugin settings → General settings → Configuration backup, choose:

Automatic backup: saving AI connections or current dictionary scripts updates _fastcm_backup.zip in the media folder, which syncs with Anki media.
Manual backup: AI connections and scripts stay local; use Import ZIP / Export ZIP explicitly, with the media folder as the default location. Export includes saved settings only, so save edits with OK first. Import warns that unsaved changes on the current settings page will be discarded.

The ZIP contains only AI connection settings and current Python dictionary scripts. Ordinary preferences, field mappings and script enable flags are saved in the media folder’s _fastcmcfg.json, including in manual backup mode. Dictionary paths stay local.

API keys are stored locally and managed by the operating system. They are not uploaded or synced with your configuration.

In automatic mode, differing settings offer Overwrite (local to backup), Sync (backup to local) or Cancel (keep both for now). Differing scripts use per-file choices, described below.

Backups exclude dictionary files and paths, dictionary JavaScript, conversations and recycled scripts. Add dictionary folders separately on another computer. On first use, current FastCard Maker settings take precedence, with legacy FastWQ settings used when absent; subsequent saves use FastCard Maker filenames.

2. Use dictionaries and parser scripts

Add dictionaries and choose data items

Open FastCM → Options → Dictionary manager → Dictionary folders and add the folder containing your local dictionaries. Subfolders are supported; keep associated MDD and resource files complete.
Under Local dictionaries, enable the required dictionaries and save. For online services, enable the corresponding items under Online dictionaries instead.
In Options, select a source field such as Word. Choose a dictionary and data item for each target, clear Ignore and save.
Query a few notes to check text and media before running a batch.

Use default parsing for the complete entry, or a custom parser for separate meanings, examples or pronunciation. Add existing .py parsers through Import script; placing Python files beside dictionaries does not import them.

New scripts prefer the MDX filename, with an optional internal title, rather than a fixed machine path. If a dictionary moves, add its new folder and rescan. Legacy fixed paths and existing audio handling remain supported; filename matching takes precedence when both bindings exist. Multiple matches require you to adjust imported folders rather than silently choosing a file.

Missing dictionaries or scripts keep their original name with a strikethrough in the manager and Options, and queries skip them. Each manager row has a Delete button that requires confirmation: missing entries are removed and existing scripts move to the local recycle bin. Options has no delete button. Reselect invalid dictionary data items when needed.

If images or audio fail to import, check MDD/resources first, then try fuzzy media matching under Dictionary settings and inspect the result. Dictionary JavaScript may be changing media paths; try AI parsing below. Missing source media cannot be recreated by a parser.

Generate or revise a parser with AI

Click AI parse beside the dictionary. Use your separate Dictionary parsing profile if available, with reasoning settings supported by its model.

Choose an AI profile. Default samples cover a noun (concept), verb (maintain) and adjective (relevant), with alternatives when missing. You can enter three different words found in the dictionary instead.
Describe the task in Command, for example “Extract meanings, phonetics and pronunciation; check JS audio-path changes and preserve other fields.” Parsing uses this command without the connection’s Skill.
Select Send HTML, Send JS and Send original script as needed. Include HTML and JS for audio-path repairs and the original parser when revising one. Missing attachments are skipped.
Choose the output mode and click Send three entries and generate. Inspect the preview and source, click Debug if needed, then Save script. Enable it in Dictionary manager and select its data items in your field mapping.

Return JSON rules is checked by default: AI describes extraction rules and the add-on generates Python, suitable for standard extraction. Uncheck it for complete Python that can handle complex paths or restructure nested content. Both modes save .py files, with matching instructions used for generation and follow-up messages.

Direct Python output receives syntax and structure checks during generation. Clicking Debug or loading a saved script executes code; passing checks does not guarantee correct results.

Read debug results and continue the conversation

Debug checks three samples and adds results and detailed errors to Information history. Enter a follow-up below and click Send message to ask for changes.

Messages, debug results and follow-up input

Include all messages is selected by default. Include checked messages sends only history checked on the right. New messages start unchecked; select individual debug reports when you want AI to consider them.
Messages show timestamps and fold after five lines. Use Expand / Collapse after a message. Check messages to Copy selected content, or use Copy all / Export Markdown.
Import conversation reads TXT or Markdown as reference context. Messages are not saved automatically; export before closing if you want to keep them.

History and source code increase token use. For long conversations, include fewer checked messages or close the window and start again. Review each AI revision and Save script to apply it.

Choose script versions and restore an earlier one

In automatic mode, differing local and backup scripts appear in a file list. Choose Local version or Backup version for each; Backup is selected by default. Header checkboxes select an entire side before individual adjustments.

Each side shows file size. Open displays read-only source without executing it. Choosing a missing side deletes that file, with confirmation. Close the window to leave choices unapplied.

To undo a script change, click Restore old script, select a version by filename and deletion time, then Restore or Open to inspect it first. Recycled scripts stay on this device, are excluded from backups and sync, and are cleaned up after seven days. A restored version participates in backup only after becoming the current script.

中文

FastCard Maker 使用说明

本插件基于 FastWordQuery（FastWQ）开发，保留原项目的作者署名与许可证。

FastCard Maker 可以用 AI 为笔记补充解释、例子和对比表格，也可以从本地 MDX 词典或网络词典中提取内容。两种方式可以分别使用，也可以为同一条笔记的不同字段选择不同来源。

以下先介绍 AI 制卡，再介绍词典查询。建议先选几条笔记试用，确认结果符合要求后再批量处理。截图中的字段名、词典和模型仅作示例。

一、使用 AI 制卡

1. 添加 AI 配置

打开 FastCM → 选项 → AI 设置：

点击 新增配置，选择 服务预设。插件内置 OpenAI（ChatGPT）、Claude、Gemini、DeepSeek 和 Qwen（通义千问）的接口地址与参数示例。
为这份配置取一个便于识别的名称。使用其他兼容服务时，选择自定义并填写自己的地址和模型名称。
填写 API 地址、API Key 和调用模型。API 地址需要包含接口路径，例如 https://api.deepseek.com/responses；预设填入的地址也可以修改。请使用所选服务支持的模型名称。
如果希望这个 AI 始终遵守某些要求，可在 Skill 中填写，例如“使用中文解释，回答简洁，并给出具体例子”。填写完成后点击确定保存。

打开 更多配置 → 自由参数（JSON） 可以设置思考强度。DeepSeek 和 Qwen 都默认使用 Responses，新建配置时预填 reasoning.effort 为 none，表示关闭思考。是否允许关闭、支持哪些档位，以具体模型为准。

这里添加的 AI 配置也可以用于词典脚本解析。建议另建一份名为“词典解析”的配置；需要分析较复杂的脚本时，可以尝试较高的思考强度。下表只列出思考相关设置，可用对应示例替换自由参数；点击服务名称可查看官方文档。

服务及接口较高思考强度的 JSON 示例OpenAI（Responses）{"reasoning":{"effort":"high"}}DeepSeek（Responses）{"reasoning":{"effort":"high"}}Qwen／百炼（Responses）{"reasoning":{"effort":"high"}}Claude（Messages，支持自适应思考的模型）{"thinking":{"type":"adaptive"},"output_config":{"effort":"high"},"max_tokens":16000}Gemini（Interactions）{"generation_config":{"thinking_level":"high"}}

这些写法需要所选模型支持，high 在不同服务中也不代表相同的思考预算。Claude 示例保留了接口要求的 max_tokens 输出上限，可按需要调整；旧模型的思考设置请查阅官方文档。较高思考强度通常会增加等待时间和 Token 用量，参数报错时可点击 ? 查看说明。

Qwen 等服务可能因地域或业务空间不同而使用不同地址，请以自己的控制台为准。

2. 选择字段并填写提示词

先在 Anki 中准备好笔记。以学习概念为例，可以建立“主题”“问题”“回答”“对比”四个字段：在前两个字段中填写“统计学”和“相关与因果有什么区别？”，后两个留给 AI 填写。

回到 FastCM → 选项，选择这份笔记所属的 笔记类型：

用左侧单选按钮把“主题”设为来源字段。
“主题”和“问题”不需要改写，保持勾选忽略。
取消“回答”的忽略，将内容来源设为刚才保存的 AI，再填写提示词。
如果还需要“对比”字段，也为它选择 AI 并填写自己的提示词。最后点击确定保存。

提示词可以引用多个字段。例如，在“回答”这一行填写：

结合{{主题}}，回答这个问题：{{问题}}
先直接回答，再举一个例子，不超过200字。

插件会把 {{主题}} 和 {{问题}} 替换成当前笔记中的内容。字段名必须与笔记中的名称完全一致；被引用的字段不要同时设为 AI 输出字段。字段引用写在这里的提示词中，不写在 Skill 中。

“对比”字段可以填写：围绕{{主题}}中的{{问题}}，用 Markdown 表格比较定义、例子和常见误区。 切换这一行使用的 AI 不会清除原有提示词。

学习单词时操作相同：来源字段填写单词，再让其他字段分别生成释义、例句或近义词。插件将结果写入已有笔记；如果卡片上看不到新增内容，需要在 Anki 的 卡片… 中把相应字段加入模板。

3. 先查询几条，再批量处理

在 Anki 浏览器中选中几条笔记，点击 FastCM → 查询选中项，检查结果是否正确。

确认后，可以使用 FastCM → 全类型查询：勾选要处理的笔记类型，为每个类型选择已保存的配置，再点击查询。

全类型查询会处理这些类型的全部笔记，不受当前牌组或浏览器选中范围限制。从浏览器打开时，会默认勾选已选笔记所属的类型；从首页打开时，默认不勾选任何类型。

如果某个字段已有内容，是否重新生成取决于该行的 跳过有值项。想重做这个字段时，取消该项；其他不需要处理的字段勾选忽略。修改配置后先点击确定保存，再开始查询。

4. 调整发送方式和卡片样式

打开 插件设置 → AI 设置，可以调整以下选项：

每张卡片一次发送：默认方式。同一笔记中使用同一 AI 配置的字段会合并为一次请求，收到回复后分别填入各字段。
每个字段分别发送：每个目标字段单独向 AI 请求结果。
解析 Markdown：把回复中的标题、列表和表格转换为卡片能显示的格式，默认开启。
启用 CSS 注入：默认关闭。需要美化表格等内容时，开启此项并勾选需要使用样式的笔记类型。点击 查看／编辑全局 CSS 可以修改共用样式。插件会在对应类型的笔记实际写入内容后添加样式，只保存设置不会立即修改卡片。

每分钟发送数、连接超时、请求超时和重试次数也在此页调整。词典设置用于调整词典线程数、匹配方式和填空格式；通用设置用于调整语言、定时制作和备份方式。不清楚某项的作用时，点击旁边的 ? 查看说明。

查询开始后，点击 后台运行 即可继续使用 Anki。从 FastCM → 任务进度 可以查看进度和结果；任务结束后点击 完成任务 收起进度栏。查询结果约每 30 秒写入一次，结束时会写入剩余结果。

关闭 Anki 会中断任务。下次启动时，如果出现恢复提示，可以点击 恢复结果，写入已经收到并缓存的内容；尚未完成的请求不会因此补回。

5. 自动处理新笔记（可选）

每个笔记类型默认使用第一份配置作为星标配置。需要更换时，在选项页点击另一份配置的星号并保存。

打开 插件设置 → 通用设置，勾选 启用定时制作，设置检测间隔、选择笔记类型，再点击确定。插件只处理启用后新增笔记的空字段；如果上一批任务还没有结束，本轮会跳过。

发现待处理笔记后，顶部会出现 10 秒倒计时。可以点击 立即运行 或 跳过本次；关闭提示也会跳过，不操作则自动开始。请保持 Anki 打开，任务结束后从 任务进度 查看结果。

6. 保存、备份与换电脑使用

修改设置后点击确定保存。如果修改后直接关闭窗口，会询问是否保存；没有修改则直接关闭。

在 插件设置 → 通用设置 → 配置保存 中选择备份方式：

自动备份：保存 AI 连接或当前词典脚本后，更新媒体文件夹中的 _fastcm_backup.zip，随 Anki 媒体同步。
手动备份：AI 连接和脚本先保存在本机，需要时使用 导入 ZIP／导出 ZIP，文件选择框默认打开媒体文件夹。导出只包含已保存配置，未保存的修改需先点击确定。导入会放弃当前设置页未保存的修改，确认框会提醒。

ZIP 中只保存 AI 配置和当前 Python 词典脚本。字段映射、脚本启用勾选等普通配置保存到媒体库的 _fastcmcfg.json，不放进 ZIP，手动备份模式下也一样。词典文件路径仍仅存本机。

API Key 保存在本地，由系统统一管理，不会随配置上传或同步。

自动备份模式下，打开选项时如果发现配置不同，可以选择覆盖（本机配置替换备份）、同步（备份读入本机）或取消（暂时保留两边）。如果是脚本不同，会逐个文件让你选择保留哪一侧，详见下文。

备份不包含词典文件、词典路径、词典附带的 JS、对话记录和回收站中的旧脚本。换电脑后，还需要添加这台电脑上的词典文件夹。首次使用时优先读取本版配置，没有时尝试读取 FastWQ 旧配置，后续保存使用 FastCard Maker 的文件名。

二、使用词典和解析脚本

1. 添加词典并选择提取内容

打开 FastCM → 选项 → 词典管理 → 词典目录，添加本地词典所在的文件夹。词典可以放在子文件夹中，配套 MDD 和资源文件需保持完整。
回到 本地词典，勾选要使用的词典并保存。使用网络词典时，在 网络词典 中勾选相应项目即可。
在选项页选择来源字段，例如“单词”；为目标字段选择词典和数据项，例如释义、例句或发音，取消忽略后保存。
先选几条笔记查询，检查内容和媒体文件，再决定是否批量处理。

需要整条词条时可以选择默认解析；需要单独提取释义、例句或发音时，选择提供这些数据项的解析脚本。已有 .py 脚本请通过 导入脚本 添加，把它放在词典文件夹旁不会自动导入。

新版脚本优先按 MDX 文件名关联词典，内部标题可以作为补充条件，不需要保存本机的绝对路径。词典移动后，重新添加所在文件夹并检测即可。旧脚本中的固定路径和音频处理逻辑仍然兼容；同时提供文件名和旧路径时优先匹配文件名。如果出现多个匹配文件，需要调整导入目录，插件不会随意选一个。

词典或脚本失效时，管理页和选项页会给原名称加上删除线，查询时跳过，不会改用其他来源。管理页每行的删除按钮需要二次确认；缺失项会移除记录，仍在本机的脚本会移入回收站。选项页不提供删除按钮。失效的数据项请重新选择。

图片或音频无法导入时，先确认 MDD 和资源文件完整，再到 词典设置 尝试开启媒体资源模糊匹配，并检查导入结果。如果词典附带的 JS 修改了媒体路径，可以尝试下面的 AI 解析方法；缺失的媒体文件本身无法靠脚本补回。

2. 用 AI 生成或修改解析脚本

在词典旁点击 AI 解析。建议使用前面单独建立的“词典解析”AI 配置；需要提高思考强度时，按模型支持的参数设置。

选择 AI 配置。默认三个样本词分别是名词 concept、动词 maintain 和形容词 relevant；找不到时会尝试备选词，也可以自行填写三个词典收录的不同词。
在命令中直接描述要求，例如：“提取释义、音标和发音，检查 JS 是否修改音频路径，保留其他字段。”这里只使用你的命令，不使用 AI 配置中的 Skill。
按需勾选 发送 HTML、发送 JS、发送原脚本。修复音频路径时建议发送 HTML 和 JS；修改已有脚本时同时发送原脚本。没有找到的附件不会发送。
选择输出方式，再点击 发送三个词条并生成。检查 提取预览 和 生成脚本，需要时点击调试。确认后点击 保存脚本，返回词典管理勾选启用，再为字段选择相应数据项。

返回 JSON 规则 默认勾选：AI 描述提取规则，再由插件生成 Python，适合常规提取。取消勾选后，AI 直接生成完整 Python，可以处理复杂路径或重新组织嵌套内容。两种方式最终都保存为 .py 脚本，首次生成和后续对话都会使用对应的提示词。

直接生成 Python 时，生成阶段只检查语法和结构。点击调试，以及保存后加载脚本时，会执行代码；检查通过不代表运行结果一定正确。

3. 查看调试结果并继续修改

点击调试会一次检查三个样本，结果和错误详情都显示在 信息历史。如果有问题，在下方输入框说明需要怎样修改，再点击 发送消息。

携带全部消息是默认选项，会把已有消息一并发送。选择 携带勾选的消息 时，只发送右侧勾选的历史消息；新消息默认不勾选。想让 AI 参考某条调试结果，也可以直接勾选它。
每条消息显示时间，超过五行的内容默认折叠，点击消息后的 展开／收起 查看。勾选后可 复制选中内容，也可以 复制全部 或 导出 Markdown。
导入对话可以选择 TXT 或 Markdown 文件，把已有对话作为参考。当前窗口的消息不会自动保存，需要保留时请在关闭前导出。

携带消息和源码会增加 Token 用量。对话过长时，可以只勾选有用的消息，或关闭窗口重新开始。AI 修改脚本后，需要再次检查并点击 保存脚本。

4. 选择脚本版本和恢复旧脚本

自动备份模式下，如果本机脚本与 ZIP 备份不同，会显示文件列表，让你逐行选择 本机版本或备份版本，默认选择备份。表头勾选框可以全选一侧，再逐行调整。

每一侧都显示文件大小，点击打开可只读查看源码，不会运行。选择显示“不存在”的一侧表示删除对应文件，确认前会再次提醒；不确定时可以关闭窗口，保留原状。

需要撤回脚本修改时，点击 恢复旧脚本，按文件名和删除时间选择版本，再点击恢复；也可以先点击打开查看。这些旧脚本只保存在本机，不同步、不进入备份 ZIP，保留七天后自动清理。 恢复后，它才作为当前脚本参与备份。

FastCard Maker（Create Flashcards with AI & Fast Word Query）

AnkiWeb

Maintenance

GitHub

Will it work on my Anki?

Similar addons

Description