Agent Compression & Data Services

Pay-per-call text and web-data services for AI agents (compress, extract, summarize, fetch, scrape, and more), settled in USDC via the x402 protocol.

Skills

• Compress messy unstructured text into a dense, token-minimal JSON object.
  Compress and minimize LLM context. Turn verbose, messy, unstructured text into a dense token-minimal JSON object that keeps every fact. Reduce token usage, shrink prompts, and cut context costs for AI agents.
  compressiontokenscontextjsoncost-savingtextllm
• Extract structured fields from raw text or HTML into clean JSON.
  Extract structured data from unstructured text or HTML. Pull caller-specified fields (names, prices, dates, invoice numbers, anything) into clean JSON. Turn documents and messy content into structured records.
  extractionstructured-datajsonparsingdatanlp
• Strip HTML, boilerplate and noise from text, returning clean plaintext.
  Clean and denoise text. Strip HTML tags, navigation, ads, cookie banners and boilerplate, returning only the meaningful human-readable plaintext. Prep scraped or messy content for an LLM.
  cleaningtexthtmldenoisepreprocessingplaintext
• Produce a dense summary of a document at a target length.
  Summarize documents and long text. Produce a dense, accurate summary at a target length (short, medium, long). Condense articles, reports, transcripts and web pages for fast downstream reasoning.
  summarizationsummarytldrcondensetextnlp
• Classify text into one of a caller-supplied set of labels, with confidence.
  Classify and categorize text. Assign input to one of your supplied labels with a confidence score. Zero-shot text classification for routing, tagging, intent detection and triage.
  classificationcategorizelabelingroutingintentnlp
• Translate text into a target language, preserving meaning and formatting.
  Translate text between languages. Convert content into any target language while preserving meaning, tone and formatting. Machine translation for multilingual agents and pipelines.
  translationtranslatelanguagemultilinguallocalizationnlp
• Analyze sentiment of text, returning a label and a -1..1 score.
  Analyze sentiment and emotion in text. Return a positive/negative/neutral label plus a -1..1 score. Gauge customer feedback, reviews, social posts and support messages.
  sentimentemotionopinionreviewsfeedbacknlp
• Extract the most important keywords / tags from text.
  Extract keywords, tags and key phrases from text. Identify the most important terms and topics for indexing, tagging, SEO and search. Keyword and keyphrase extraction.
  keywordstagskeyphrasestopicsseonlp
• Named-entity recognition: people, orgs, places, dates, amounts from text.
  Named-entity recognition (NER). Extract people, organizations, locations, dates and monetary amounts from text into structured JSON. Identify entities for knowledge graphs and data enrichment.
  nerentitiesnamed-entityextractionpeopleorganizations
• Detect and mask PII (emails, phones, names, IDs) in text.
  Detect and redact PII from text. Mask emails, phone numbers, names, addresses and government/card IDs with typed placeholders. Anonymize and sanitize data for privacy and compliance.
  piiredactionprivacyanonymizecompliancesecurity
• Repair malformed/truncated JSON into a valid JSON object.
  Repair broken JSON. Fix malformed, truncated or loosely-formatted JSON (single quotes, trailing commas, cut-off output) into valid parseable JSON. Recover LLM output that failed to parse.
  jsonrepairfixparsingvalidationllm-output
• Convert HTML or messy text into clean, well-structured Markdown.
  Convert HTML to Markdown. Turn web pages, HTML or messy text into clean, well-structured Markdown (headings, lists, links, tables). Make web content LLM-ready.
  markdownhtmlconversionformattingwebllm-ready
• Detect the language of text, returning name, ISO code and confidence.
  Detect the language of text. Return the language name, ISO 639-1 code and a confidence score. Language identification for routing and multilingual pipelines.
  language-detectionlanguageidentifyisomultilingualnlp
• Flag unsafe content by category (hate, violence, sexual, self-harm, etc.).
  Moderate content for safety. Flag unsafe text by category (hate, harassment, violence, sexual, self-harm, illegal). Content moderation and safety filtering for user input and generated text.
  moderationsafetycontent-filtertrust-safetytoxicitynlp
• Rewrite text in a target style or tone (formal, casual, concise, etc.).
  Rewrite and rephrase text in a target style or tone (formal, casual, concise, friendly, professional). Paraphrase, adjust voice, and polish copy while preserving meaning.
  rewriteparaphraserephrasetonestylecopywriting
• Fetch a URL and return its main content as clean Markdown or plaintext.
  Fetch and read any web page. Give a URL, get back the main content as clean Markdown or plaintext with nav/ads/boilerplate removed. Web access and browsing for sandboxed agents that can't reach the internet. Scrape a URL to LLM-ready text.
  webfetchscrapeurlbrowsingweb-accessmarkdowncrawler
• Fetch a URL and extract caller-specified fields into structured JSON.
  Scrape structured data from any web page. Give a URL and the fields you want; get clean structured JSON back. Turn websites into structured records — prices, titles, contacts, listings, specs — for sandboxed agents that can't browse.
  web-scrapingscrapeurlstructured-dataextractionwebcrawlerjson
• Detect prompt-injection and jailbreak attempts in untrusted input.
  Detect prompt injection and jailbreaks. Scan untrusted user input for attempts to manipulate an AI agent — instruction overrides, system-prompt extraction, jailbreaks, data exfiltration, tool abuse — before your agent acts on it. A security guardrail for LLM and agent pipelines.
  securityprompt-injectionjailbreakguardrailagent-securityllm-securitysafety

How to call

https://agent-compression-biz-production.up.railway.app