Skip to content

Cheaper web scraping for AI agents.

Stop paying your model to scrape. Clean web data, at scale, for a fraction of the cost.

Powering open models with bandwidth.

Terabyte-scale web data for open source models, over a residential network, for a fraction of the cost.

❯_runner-ai · claude code
❯ Scrape example.com as seen from Germany and clean it

● I'll use the Runner AI scrape skill to fetch example.com from Germany and clean it.

● Skill(runner-ai:scrape)
  └ Successfully loaded skill

✓ exit acquired · DE · 38 ms
✓ clean content · 448 B · status 200

● Done. Routed through a residential node in Germany (DE), noise tags stripped out.
Result: a minimal placeholder page (448 B · 0 credits used).
- Title: Example Domain
- Link: Learn more → iana.org/domains/example

✱ Worked for 21s · no API key
Built forClaudeChatGPTGemini
Public Data for Your Open-Source Model Today
  • Kimi
  • GLM
  • DeepSeek
  • Qwen
  • Llama
  • Mistral

AI systems need current web data

Models learn from collected data. Web access brings them new information when they need it.

Training and answers

Web data supports AI before and after training.

  • OpenAI
  • Anthropic
  • Gemini
  • Perplexity
Learning from web content
Collected web content provides material to train and improve AI models.
Answering with current facts
Live sources provide information that has changed since the model was trained.

Open-source models

Teams can run and adapt these models themselves.

  • Kimi
  • DeepSeek
  • Qwen
  • Mistral
Knowledge for a specific purpose
Teams use selected web content to build AI tailored to their business.
A connection to the live web
Developers add search or scraping tools so the model can retrieve new information.

Agent workflows

Agents use web data while carrying out tasks.

  • Cursor
  • Zapier
  • n8n
  • Browserbase
Research before action
Agents can gather relevant information from websites before deciding what to do.
Ongoing monitoring
Agents can check websites regularly and report important changes without repeated manual research.

Scrape globally. Get local context.

Pin any request to a country and read the page the way a local does, through a real residential connection there. Unavailable countries fail loudly — never a silent fallback that still bills you.

—

Locations in the catalog

Loading locations

—

Country coverage

Country coverage pending

Including
  • United States
  • United Kingdom
  • Germany
  • France
  • Brazil
  • Japan
and many more
    6,000+

    US cities covered

    20,000+

    cities covered globally

    1,600+

    states & regions covered

    City and regional figures as of 15 October 2026.

    34 data types. 7 categories. One flat rate.

    Across media, documents and web data — priced by the byte, not the page.

    $0.05/ GB

    of successfully delivered output

    • Same rate for every data type
    • No per-page fees
    • No credit multipliers
    • Failed requests never charged

    Video

    9 types

    Clips, recordings & video containers

    • MP4
    • WEBM
    • MKV
    • AVI
    • MOV
    • +4and more3GP, FLV, OGV, WMV

    Audio

    7 types

    Speech, music & lossless audio

    • MP3
    • OGG
    • WAV
    • FLAC
    • OPUS
    • +2and moreM4A, AAC

    Images

    7 types

    Raster, vector & animated images

    • JPEG
    • PNG
    • WEBP
    • SVG
    • GIF
    • +2and moreBMP, TIFF

    Documents

    5 types

    Documents & spreadsheets

    • PDF
    • DOCX
    • XLSX
    • ODT
    • RTF

    Web pages

    2 types

    Page source & readable text

    • HTML
    • Plain text

    Structured data

    3 types

    Objects, markup & tables

    • JSON
    • XML
    • CSV

    Archives

    1 type

    Compressed file delivery

    • GZIP

    Page outputs

    4 formats

    Every web page, returned the way your pipeline wants it

    Raw HTML
    Original page source
    Clean HTML
    Readable page content
    Markdown
    Text for AI workflows
    Links
    Extracted page links

    What a flat rate does to the bill

    One flat rate for delivered bytes, without per-page fees or credit multipliers. Per-page pricing punishes small pages hardest — and most pages are small.

    Estimated cost per GB

    USD per GB, calculated by page size across four plans per competitor.

    Page sizeRunnerDecodoFirecrawl
    10 KiB$0.05$20.26–$48.83$58.50–$312.50
    50 KiB$0.05$4.05–$9.77$11.70–$62.50
    100 KiB$0.05$2.03–$4.88$5.85–$31.25
    500 KiB$0.05$0.41–$0.98$1.17–$6.25
    1 MiB$0.05$0.20–$0.48$0.57–$3.05
    5 MiB$0.05$0.040–$0.095$0.114–$0.610
    10 MiB$0.05$0.020–$0.048$0.057–$0.305

    Assumes all included requests are used. Firecrawl uses annual pricing at one credit per page.

    Decodo’s $249 plan costs less above 4.15 MB per page.

    Estimated cost per GB of source files

    Four plans per competitor. Costs use source-file sizes. Outputs differ.

    File type & sizeRunnerDecodoFirecrawl
    Thumbnail 3.95 kB$0.05$52.54–$126.61$151.68–$810.33
    Icon / graphic 17.75 kB$0.05$11.69–$28.17$33.74–$180.26
    Web photo 102.40 kB$0.05$2.03–$4.88$5.85–$31.25
    Audio clip 563.76 kB$0.05$0.37–$0.89$1.06–$5.68
    Document 2.22 MB$0.05$0.094–$0.226$1.35–$7.22
    Video clip 2.85 MB$0.05$0.073–$0.176$1.05–$5.62
    Longer video 6.69 MB$0.05$0.031–$0.075$0.45–$2.39
    Lossless audio 12.36 MB$0.05$0.017–$0.041$0.24–$1.29

    Assumes all included requests are used and Firecrawl is billed annually. Providers return different outputs.

    Firecrawl’s five-credit charge above 768 kB is an unverified assumption.

    Estimated equivalent cost in USD per GB of delivered payload, modeled from each provider’s published plans at full allowance use — not measured bills.

    One flat rate, one prepaid balance that never expires.

    Ways to access Runner

    The API is live today. Agent tools and platform integrations extend the same access layer.

    Direct API

    Live

    Call Runner directly for scrape requests, country routing and account-managed usage.

    • REST scrape requests and geo controls
    • Console accounts and API keys
    • Credits and receipts

    Call from your application

    • TypeScript
    • Python
    • cURL

    MCP server

    Planned

    Agents will call Runner as a native tool for natural-language scrape tasks.

    • Natural-language scrape and extract
    • Agent workflows and tool calls
    • Scoped Runner credential

    Agent tools

    • Claude
    • Cursor
    • ChatGPT

    Platforms & SDKs

    Planned

    Typed clients, automation tools and data destinations will bring Runner into existing AI workflows.

    • TypeScript and Python SDKs
    • AI frameworks and automation tools
    • S3, GCS, Postgres and IPFS destinations

    Planned ecosystem paths

    • LangChain
    • LlamaIndex
    • Vercel AI SDK
    • n8n
    • Zapier
    • Make
    • Dify
    • AWS
    • Google Cloud
    • Postgres
    • IPFS

    One credential. One billing surface.

    One API for web data. Get your key.

    Request access, add credit, and start pulling clean, LLM-ready data. No contracts, no minimums, no sales call required.

    The data you need, at a price no one can match.

    A terabyte of web data for the cost of lunch. Book a call and we will scope your volume, geolocation, and delivery.