Custom AI Actions - Build Your Own Reading Tools Inside Read Frog

Custom AI Actions Turn Read Frog Into Your Own AI Workspace

Translation is only one part of reading. Sometimes you want an article rewritten in simpler language. Sometimes you want a quick summary, a vocabulary list, or a structured response you can copy somewhere else.

That is exactly what Custom AI Actions are for.

With Custom AI Actions, you can define your own prompts inside Read Frog and run them directly from the selection toolbar. Pick text on any page, trigger your action, and get a result shaped for the task you actually care about.

What You Can Build

Custom AI Actions are flexible enough for both quick utilities and repeatable workflows:

• Summarize a long paragraph in three bullets
• Explain a difficult sentence in plain English
• Rewrite text in a more natural tone
• Extract key terms and definitions
• Turn notes into JSON-like structured output
• Create study cards, outlines, or action items from selected text

If you already use different models for different jobs, this feature makes that workflow much faster.

Where to Configure It

In the current extension code, Custom AI Actions live on their own settings page:

1. Open the Read Frog options page.
2. In the left sidebar, open Custom AI Actions.
3. Make sure you already enabled at least one LLM provider.

That last step matters. The settings UI will block creation if there is no enabled LLM provider, because this feature depends on structured AI output.

What Templates Exist Right Now

The add-action dialog currently ships with 3 templates:

• Dictionary: looks up a term and returns fields like term, phonetic, part of speech, definition, paragraphs, paragraph translation, and CEFR difficulty.
• Improve Writing: analyzes grammar and wording issues, then returns an error analysis plus an improved version.
• Blank: starts from scratch with a single default result field.

There is also a built-in default action in the extension config: Dictionary is pre-created in DEFAULT_CONFIG.selectionToolbar.customActions.

So if you install the feature and never add anything manually, the product already treats dictionary lookup as the first Custom AI Action.

What You Actually Configure

The Custom AI Actions editor currently exposes these fields:

• Name
• Icon (Iconify)
• Provider: must be an enabled LLM provider that supports structured output
• System Prompt
• Prompt
• Output Schema

The output schema is the most important part. Each field in the schema has:

• Field Name
• Field Type: currently string or number
• Description
• Enable speaking

At least one output field is required, and field names must be unique.

Which Prompt Tokens You Can Use

The code currently supports 4 insertable tokens in both the system prompt and prompt:

• {{selection}}: the exact selected text
• {{paragraphs}}: intersecting paragraph text joined with blank lines
• {{targetLanguage}}: the user's current target language
• {{webTitle}}: the page title

One useful implementation detail from the source: {{paragraphs}} is truncated to the first 2000 characters before being sent to the custom action request.

What the Result Looks Like

Custom AI Actions do not return a loose chat blob. The extension builds a structured output contract from your schema and asks the model to return exactly one JSON object.

Then Read Frog renders that object back into field rows inside the selection toolbar result panel:

• string and number fields are shown separately by type
• each row uses your field name as the label
• if you enabled speaking for a field, that field gets a speak button

This is the part that makes the feature much more reusable than a generic prompt box.

A Practical Way to Create Your First Action

If you want a good first custom action, use Blank and fill it like this:

Name

Reading Summary

System Prompt

You are a concise reading assistant.
Summarize the selected content using the required structured output.
Prefer concrete wording over vague abstraction.

Prompt

Selection: {{selection}}

Paragraphs: {{paragraphs}}

Target language: {{targetLanguage}}

Webpage title: {{webTitle}}

Output Schema

• summary -> string
• keywords -> string
• nextSteps -> string

This gives you a repeatable reading helper that works on articles, docs, and tutorials without rewriting the same prompt every time.

Source-Based Recommendation

If you are new to the feature, do not start by inventing a huge workflow.

Start with one of the shipped templates:

• choose Dictionary if you read in another language and want vocabulary help
• choose Improve Writing if you edit your own text often
• choose Blank if you already know the exact structure you want back

That matches the current product design much better than treating Custom AI Actions like an empty AI sandbox. The extension is clearly optimized for small, repeatable, structured actions that run directly from the selection toolbar.