Agent-native
Store or update an item in agent memory with its text, namespace, and metadata for later semantic retrieval. Backed by Pinecone serverless.
Retrieve the top-k most similar items from agent memory for a query, with optional namespace and metadata filter. Backed by Pinecone serverless.
Delete one or more items from agent memory by id, with optional namespace. Backed by Pinecone serverless.
Run Python, Node, or Bash code in an isolated sandbox with optional files, env, and network; returns output. Dynamic pricing: ceil(timeout_ms × 10000 / 60000), ×2 if allow_network=true. Powered by Modal.
Load a web page in an isolated worker and return a screenshot, DOM, or both, with optional interactions and viewport. Powered by Playwright.
Critique a target response against a prompt with multiple LLMs in parallel and return a synthesized assessment. Claude, GPT, and Gemini.
Answer a question by polling up to 5 LLMs in parallel and returning a consensus (majority, weighted, or unanimous), with optional explanation.
Prepaid research-agent bundle ($0.50) — 200× url-text + 100× sitemap-parse + 40× nlp/sentiment.
Prepaid document-agent bundle ($0.30) — 50× ocr + 20× md-pdf + 15× memory/store.
Prepaid voice-agent bundle ($1.00) — 100× whisper + 3× audio/voice-clone + 1× video/lipsync.
Documents
Convert PDF to Markdown; returns { markdown, pages }. Priced per page. Uses pdf2md.
Extract plain text from PDF; returns { text, pages }. Priced per page. Uses pdf-parse.
Extract structured per-page text blocks from PDF; returns { pages }. Priced per page. Uses pdf-parse + heuristics.
Extract invoice fields (vendor, date, amount, VAT, items) from PDF or image. Uses tesseract + regex/LLM.
OCR business card image to vCard fields (name, email, phone, company). Uses tesseract + parsing.
Detect and extract table from image to CSV rows/columns. Priced per image. Uses tesseract + table-detect.
Convert DOCX upload to Markdown; returns { markdown }. Uses mammoth.
Parse XLSX workbook to JSON sheets with row arrays. Uses xlsx (SheetJS).
OCR handwritten text from image; returns { text, confidence }. Uses tesseract handwriting LSTM.
Merge multiple PDFs into one file. Dynamic: base $0.003 (≤10 pages summed) + $0.0003/page extra. Uses pdf-lib.
Split PDF into separate page files. Dynamic: base $0.003 (≤10 pages) + $0.0003/page extra. Uses pdf-lib.
Compress PDF file size via Ghostscript. Dynamic: base $0.005 (≤10 pages) + $0.0005/page extra.
Render HTML body to PDF. Uses playwright print.
Render Markdown to PDF. Uses markdown-it + playwright.
Extract PPTX slides to structured text/slide list. Uses pptxgenjs reverse.
Extract text from an image with layout and handwriting support using a vision model. Powered by GLM-4V-9B via Replicate.
Audio
Transcribe audio to text with segments; priced per minute. Uses whisper.cpp small.
Transcribe audio then produce summary and key points; priced per minute. Uses whisper + Claude Haiku.
Detect spoken language from audio sample; returns { language, confidence }. Uses whisper detect.
Speaker diarization with timed segments; priced per minute. Uses pyannote-audio.
Extract audio from video and transcribe; priced per minute. Uses ffmpeg + whisper.cpp.
Extract PNG thumbnail frame from video at given second. Dynamic: base $0.0015 (≤30s) + $0.0005/min extra. Uses ffmpeg.
Text-to-speech to MP3 from { text, voice, lang }; priced per 500 characters. Uses piper-tts.
Convert audio file to requested output format. Uses ffmpeg.
Extract audio track from video as MP3. Dynamic: base $0.002 (≤60s) + $0.001/min extra. Uses ffmpeg.
Convert video clip to animated GIF. Dynamic: base $0.003 (≤30s) + $0.0015/min extra. Uses ffmpeg.
Loudness-normalize audio (EBU R128 loudnorm). Uses ffmpeg loudnorm.
Convert SRT subtitles to WebVTT format. Uses pure JS parsing.
Transcribe audio from a URL to text using a premium model. Dynamic: base $0.005 (≤60s) + $0.0025/min extra. Powered by Deepgram nova-2.
Estimate the tempo (beats per minute) of an audio file. Powered by librosa.
Estimate the musical key of an audio file using the Krumhansl-Schmuckler method. Powered by librosa.
Generate a short video from a text prompt (optional reference image, 16:9 or 9:16). Powered by Google Veo 3 Fast on Replicate.
Synchronize a video's lip movements to a provided audio track, with selectable sync mode. Powered by Sync Labs lipsync-2-pro on Replicate.
Synthesize speech in a cloned voice from a reference sample and target text (selectable language). Powered by Minimax voice-cloning on Replicate.
Transcribe audio to text with a premium model; priced per minute. Powered by OpenAI gpt-4o-transcribe via Replicate.
Generate music from a text prompt (optional negative prompt and seed). Powered by Google Lyria 2 via Replicate.
Images
OCR image text with confidence and language (fra+eng+spa); priced per page. Uses tesseract.js.
Generate natural-language caption for image. Uses BLIP-2 or external API.
Classify image content tags with scores. Uses yolov8 + classifier.
Extract dominant color palette (hex) from image. Uses sharp + quantize.
Read EXIF metadata from image file. Uses exifr.
Score image for NSFW content; returns { nsfw_score, safe }. Uses nsfwjs (TensorFlow.js).
Detect objects with labels, bounding boxes, and scores. Uses yolov8 ONNX.
Count faces in image (no identity). Uses mediapipe face detect.
Resize image to requested width/height. Uses sharp.
Compress image with configurable quality. Uses sharp.
Convert image to requested output format. Uses sharp.
Decode QR code from image to payload string. Uses jsQR.
Encode data string as QR code PNG image. Uses qrcode.
Decode 1D/2D barcode from image. Uses zxing-wasm.
Generate barcode image from data and format. Uses bwip-js.
Embed image into single-page PDF. Dynamic: base $0.002 (≤4 MP) + $0.0005/MP extra. Uses pdf-lib.
Encode uploaded image as base64 string. Native implementation.
Remove image background to transparent PNG. Uses rembg (u2net).
Generate an image from a text prompt with configurable size, steps, and guidance. Powered by SDXL on Replicate.
Generate an image from a text prompt with configurable aspect ratio, format, and quality. Powered by FLUX 1.1 Pro on Replicate.
Generate an image from a text prompt with configurable aspect ratio, format, and safety level. Powered by Google Imagen 4 on Replicate.
Edit an image from a text prompt, returning JPG or PNG. Powered by Google nano-banana (Gemini 2.5 image editor) on Replicate.
Edit an input image from a text prompt with configurable aspect ratio and output format. Powered by FLUX Kontext Max on Replicate.
Generate a CLIP embedding vector for an image. Powered by clip-vit-large-patch14 on Replicate.
Remove an image's background to a transparent PNG with refined hair and edge handling. Powered by Bria via Replicate.
Restore and upscale faces in a photo with configurable upscale and fidelity. Powered by CodeFormer via Replicate.
Generate a textured 3D asset from a single image (configurable texture size and mesh simplification). Powered by Trellis via Replicate.
Web
Fetch URL and extract main article text (Readability, no JS). Uses mozilla/readability.
Fetch Open Graph and page metadata (title, description, og_image). Uses unfurl.js.
List hyperlinks extracted from fetched URL HTML. Uses cheerio.
Follow redirect chain for URL; returns ordered hop list. Uses fetch chain follow.
Check URL reachability with HEAD request; returns alive, status, latency. Uses fetch HEAD.
Capture full-page PNG screenshot of URL at given viewport. Uses playwright.
Render URL to PDF via headless browser print. Uses playwright print.
Parse sitemap.xml URL list from sitemap document. Uses xml2js.
Parse robots.txt rules and sitemap references. Uses pure JS parsing.
Parse RSS/Atom feed URL to JSON items list. Uses rss-parser.
Extract plain text from HTML body. Uses turndown / cheerio.
Extract hyperlinks from provided HTML. Uses cheerio.
Download site favicon as PNG from URL. Uses favicon-fetcher.
Return HTTP response headers from URL HEAD request. Uses fetch HEAD.
Extract HTML page title from URL. Uses cheerio.
Search the web and return ranked results with snippets (max_results, basic/advanced depth). Powered by the Tavily API.
Run Python 3 code in an isolated sandbox with optional stdin and args; returns stdout, stderr, and exit code. Powered by self-hosted Piston.
Run Node.js code in an isolated sandbox with optional stdin and args; returns stdout, stderr, and exit code. Powered by self-hosted Piston.
Fetch a JS-rendered page and extract its HTML or text, optionally scoped to a CSS selector. Powered by Playwright stealth.
Translate text into a target language, with optional source language. Powered by the DeepL Free API.
Store a key-value entry with optional semantic embedding and TTL (default 30 days). Backed by SQLite with ONNX embeddings.
Retrieve stored entries by exact key or by semantic similarity (top-k). Backed by SQLite with all-MiniLM-L6-v2 embeddings.
Fetch a package's README/documentation by name and ecosystem (npm, pypi, github), with optional section. Cached 1h via GitHub raw.
Classify the sentiment of English text (positive/negative with score). Runs DistilBERT ONNX locally on CPU via transformers.js.
Detect named entities (people, organizations, places, dates) in English text. Powered by spaCy via a Python runner.
Generate a 384-dimension embedding vector for input text, optionally normalized. Powered by all-MiniLM-L6-v2.
Run a multi-step browser flow (click, fill, wait, scroll, evaluate; max 5 steps) in a persistent session. Powered by Playwright stealth.
Scrape a JS-rendered page; returns rendered HTML, extracted text, and optional per-selector fields. Powered by Playwright stealth.
Parse a manifest (package.json, requirements.txt, pyproject.toml, or Cargo.toml) into a structured dependency graph. Local parser.
Convert an address or place name into latitude/longitude coordinates, with optional country filter. Powered by Nominatim (OpenStreetMap).
Compute a route between two coordinates with distance, duration, and geometry (driving, cycling, or walking). Powered by OSRM.
Detect and mask PII (email, phone, IBAN, IP, credit card, SSN, URL) in text, with optional offsets. Local regex and NER.
Produce an extractive summary of text (configurable sentence count, en/fr/auto). Uses LexRank locally on CPU, no LLM.
Run a full-text search across stored memory entries within a namespace (configurable limit). Powered by SQLite FTS5.
Run a Bash script in an isolated sandbox with optional stdin and args; returns stdout, stderr, and exit code. Powered by self-hosted Piston.
Generate a chat completion from a prompt (optional system prompt, max_tokens, temperature). Powered by OpenAI GPT-4o mini via Replicate.
Generate a chat completion from a prompt (optional system prompt, max_tokens, temperature). Powered by Anthropic Claude 3.7 Sonnet via Replicate.
Generate a chat completion from a prompt (optional system prompt, max_tokens, temperature). Powered by DeepSeek V3 via Replicate.
Generate a reasoning-model completion from a prompt (optional system prompt, max_tokens, temperature). Powered by DeepSeek R1 via Replicate.
Validation
Validate email syntax, MX, disposable/role flags. Uses mailcheck + DNS MX.
Validate and format phone number internationally. Uses libphonenumber-js.
Lookup carrier, country, and line type for phone number. Uses libphonenumber + lookup table.
Resolve BIC, bank name, and country from IBAN. Uses iban-bic table.
Validate IBAN checksum and country code. Uses iban.js.
Fetch French company record by SIREN from INSEE Sirene API.
Fetch French establishment record by SIRET from INSEE Sirene API.
Validate EU VAT number via VIES; returns name and address when available.
Geolocate IP to country, city, latitude, longitude. Uses maxmind GeoLite2.
Lookup ASN and organization for IP address. Uses maxmind ASN DB.
Query domain WHOIS (registrar, dates, nameservers). Uses whois-json.
Resolve DNS records for domain and record type. Uses node:dns.
Inspect TLS certificate for domain (issuer, validity, SANs). Uses tls socket peek.
List MX mail exchanger records for domain. Uses node:dns.
Hash password with bcrypt, argon2, or scrypt per algo parameter. Uses bcrypt + argon2.
Score password strength with feedback. Uses zxcvbn.
Check URL against Google Safe Browsing threat lists. Uses Google Safe Browsing API.
Test regex against input; returns matches and capture groups. Native implementation.
Validate JSON data against JSON Schema; returns errors if invalid. Uses ajv.
Extract values from JSON document using JSONPath expression. Uses jsonpath.
Get the latest quote for a stock symbol (price, change, volume), cached 30s. Powered by Alpha Vantage.
Build or verify an AP2 Intent Mandate as EIP-712 typed data for agent identity (spend cap, expiry, wallet). Powered by viem.
Screen a name or wallet against the EU consolidated sanctions list (CFSP), with optional fuzzy matching and threshold.
Get a company overview by symbol (sector, market cap, financial ratios). Powered by Alpha Vantage OVERVIEW.
Settlement audit trail — Ed25519-signed PDF (tx_hash, facilitator, endpoint, amount, signature). POST /v1/_audit/:txHash.
Conversions
Convert Markdown string to HTML. Uses markdown-it.
Convert HTML to Markdown. Uses turndown.
Convert JSON object to YAML string. Uses js-yaml.
Parse YAML string to JSON. Uses js-yaml.
Convert JSON to XML string. Uses xml-js.
Parse XML string to JSON. Uses xml2js.
Parse CSV text to JSON rows with optional delimiter. Uses csv-parse.
Serialize JSON rows to CSV string. Uses csv-stringify.
Encode text or binary data to base64. Native implementation.
Decode base64 string to raw data. Native implementation.
Percent-encode text for URL query components. Native implementation.
Percent-decode URL-encoded text. Native implementation.
Convert hex string to base64-encoded bytes. Native implementation.
Convert base64-encoded bytes to hex string. Native implementation.
Format Unix timestamp with locale, format string, and timezone. Uses date-fns-tz.
Convert hex color to RGB components. Native implementation.
Convert RGB color to HSL components. Native implementation.
Generate placeholder lorem ipsum text by paragraphs or words. Uses lorem-ipsum.
Generate random UUID (v4). Uses crypto.randomUUID.
Generate URL-safe slug from text. Uses slugify.
Community
IronPrompt — anti-hallucination prompt compiler for Claude. Plain-language task in, structured multi-phase prompt out.
Safe no-payment x402 and agent-commerce readiness snapshot with proof links and paid review path.