AI Translator, Summarizer & Rewriter

Transform any URL or raw text through five AI-powered modes: translate, summarize, rewrite, extract-key-points, and tldr. URL inputs are automatically fetched and extracted to clean text before processing. Every dataset row includes token counts and vendor cost, so you know exactly what each operation cost.

What it does

• Translate — Render text in any language. Provide an ISO 639-1 code or a language name.
• Summarize — Condense content in formal, casual, technical, marketing, or SEO style.
• Rewrite — Reword content in a chosen style while preserving length and meaning.
• Extract key points — Produce a bulleted list of the core ideas.
• TL;DR — One-paragraph summary of any length content.

URL inputs go through a full fetch + article extraction pipeline to strip navigation, ads, and boilerplate before the AI call. Raw text inputs are processed directly.

Input

Field	Required	Description
inputs	Yes	Array of objects. Each must have url or text (or both). Optional language field for source language hint.
openrouterApiKey	Yes	Your OpenRouter API key (sk-or-...). Get a free key at openrouter.ai.
mode	No	translate (default), summarize, rewrite, extract-key-points, tldr.
targetLanguage	Required when mode=translate	ISO 639-1 code (e.g. es) or language name (e.g. French).
style	No	formal (default), casual, technical, marketing, seo. Applies to summarize and rewrite modes.
model	No	OpenRouter model ID. Default: openai/gpt-4o-mini. Options: openai/gpt-4o, anthropic/claude-3.5-haiku, anthropic/claude-3.5-sonnet, google/gemini-flash-1.5, mistralai/mistral-small.
maxOutputTokens	No	Maximum response length per input. Default: 1024.
useCache	No	Cache results by input hash (7-day TTL). Identical inputs return instantly without consuming tokens. Default: true.
maxItems	No	Cap on number of inputs to process. Default: 15.

Example input

{
  "inputs": [
    { "url": "https://example.com/article" },
    { "text": "The quick brown fox jumps over the lazy dog." }
  ],
  "mode": "translate",
  "targetLanguage": "es",
  "model": "openai/gpt-4o-mini",
  "openrouterApiKey": "sk-or-your-key-here"
}

Output

Each input item produces one dataset row:

Field	Type	Description
input	object	Original input item
mode	string	Operation mode used
model	string	OpenRouter model ID
output	string	Transformed content
inputTokens	number	Tokens consumed by the prompt
outputTokens	number	Tokens consumed by the completion
costUsd	number	Vendor cost in USD for this record
cached	boolean	True when served from local cache
extractedFromUrl	boolean	True when a URL was fetched and extracted
sourceLanguage	string	Source language
targetLanguage	string	Target language (translate mode only)
status	string	success or error
errorMsg	string	Error detail on failure

Example output row

{
  "input": { "url": "https://example.com/article" },
  "mode": "translate",
  "model": "openai/gpt-4o-mini",
  "output": "El rápido zorro marrón salta sobre el perro perezoso.",
  "inputTokens": 312,
  "outputTokens": 298,
  "costUsd": 0.00000918,
  "cached": false,
  "extractedFromUrl": true,
  "sourceLanguage": "en",
  "targetLanguage": "es",
  "status": "success",
  "errorMsg": null
}

OpenRouter API key

This actor calls the OpenRouter API using your own key. OpenRouter provides a unified gateway to 200+ LLM models from OpenAI, Anthropic, Google, Mistral, and others.

1. Sign up for free at openrouter.ai
2. Create an API key in your dashboard
3. Add credits (pay-as-you-go starting from a few dollars)
4. Paste the key into the openrouterApiKey field

The actor passes your key directly to the OpenRouter API. No keys are stored. Costs are charged to your OpenRouter account balance.

Model selection

Model	Best for	Cost
openai/gpt-4o-mini	Fast, cheap, good quality (default)	Very low
google/gemini-flash-1.5	Fastest option	Very low
mistralai/mistral-small	European languages	Low
anthropic/claude-3.5-haiku	Nuanced rewrites	Low
openai/gpt-4o	Highest quality	Medium
anthropic/claude-3.5-sonnet	Complex content	Medium

For most translation and summarization tasks, gpt-4o-mini gives excellent results at minimal cost.

Caching

When useCache is enabled (default), results are cached using a SHA-256 hash of the input text, mode, target language, model, and style. A cache hit returns the stored output immediately with zero LLM calls. The cache TTL is 7 days.

Concurrency and rate limits

The actor processes up to 10 inputs in parallel. On rate limit responses (HTTP 429), the actor backs off exponentially with jitter and retries up to 5 times before recording an error for that item. Remaining inputs continue processing.

Error handling

Failed inputs produce a dataset row with status: "error" and an errorMsg. The actor continues processing remaining inputs. Common error causes:

• Invalid or missing API key
• URL unreachable or extraction returned empty content
• OpenRouter model quota exceeded
• targetLanguage missing when mode is translate