
Reworkd
Reworkd is an end-to-end web data extraction platform for data teams: LLM agents read each page, write the scraping code themselves and repair it when the site changes. Its publisher has announced the product is being sunset.
What is Reworkd?
Reworkd is an end-to-end web data extraction platform published by Reworkd AI, Inc., a company incorporated in California and based in San Francisco. The home page states the pitch bluntly: "Effortlessly extract web data at scale. No code. No maintenance. No worries." The documentation puts the mechanism more plainly still, explaining that Reworkd uses LLMs to parse, understand and interact with web pages so that users can scrape web data at scale. What makes the approach distinctive is that the agents do not read values off a page and hand them back. They read the page and then write the extraction code, and that code is what the platform runs. Reworkd presents this as its answer to hallucination: generated code either works or fails loudly, where a model asked to guess a value could simply invent one. Around it sit three further claims from the home page, namely automated extraction, self-healing scrapers that notice when a site changes and repair themselves, and support for any data type, text, images and documents alike. The product model is easy to describe. A Group gathers source URLs that share one schema and one scraping frequency. Each source URL is a Job, configurable with its own proxy type and timeouts. Jobs break down into Stages that distinguish category pages, listing pages and detail pages, and every execution is a Run. The generated code leans on Harambe, Reworkd's own open-source SDK built on Playwright. The company came out of AgentGPT, one of the first generalist autonomous agents, which collected around 25,000 GitHub stars and more than 500,000 users within months before the founders pivoted to web extraction. The home page counts 533,029,180 rows of data extracted and names Paul Graham, Nat Friedman and Daniel Gross among its backers, while the seed announcement framed the ambition as building the unified API for the internet, starting with read access. One thing has to be said plainly: a banner across the whole site announces that the product is being sunset on February 6, 2025, with a migration contact at Srijan@reworkd.ai, and the application's login screen now reports that no login options are enabled.
What it does
- Scan a target website and automatically generate the extraction code for it
- Run extractors at scale across hundreds or thousands of source sites
- Validate every saved row against a user-defined data schema
- Detect a change in a site's structure and repair the scraper automatically
- Handle pagination, infinite scroll, dynamic content, retries and rate limiting
- Solve captchas and get past anti-bot protections through managed proxies
- Schedule runs per group, download linked files and export results as JSON, CSV or through the API
When to use Reworkd / When not to
A quick filter to help you decide if Reworkd is the right fit.
When to use Reworkd
- Data teams that need to extract structured data from hundreds or thousands of different websites
- Engineering teams tired of rewriting scrapers every time a target site changes its layout
- AI and machine learning teams assembling training or fine-tuning corpora from public web sources
- Product teams whose offering is data-constrained and depends on continuously refreshed web data
- Companies that want managed scraping infrastructure without ever touching a headless browser or a residential proxy
When not to use Reworkd
- Teams looking for a production tool to rely on long term, since the product has been announced as sunset and the application login is disabled
- Organisations that require a signed DPA, a published sub-processor list or a security certification such as SOC 2 or ISO 27001
- Users who need a documented way to keep their data out of model training
- Companies with a self-hosting or on-premise requirement, since the whole solution is managed on Reworkd's side
- Non-English speakers and anyone expecting a mobile app or a browser extension, as the product is web-only and English-only
How to use Reworkd
A typical end-to-end flow, from setup to results.
- Create a Group that gathers the source URLs sharing one schema and one scraping frequency
- Define the schema describing the structured data you want back; every job in the group uses the same one
- Add your source URLs, each of which becomes a Job you can configure with its own proxy type and timeouts
- Let the agents scan the pages and generate the extraction code, which runs through the Harambe SDK
- Follow how a job breaks into stages, category page, listing page and detail page, with child jobs queued by the source job
- Launch a Run, track its status and replay it if it fails
- Set the cadence in the group settings tab, overriding it on an individual job when a source needs a different rhythm
- For linked documents, add a URL-type field to the schema and switch on Download file from URL
- Pull results out with bulk JSON or CSV exports from the interface, or through the API, which the documentation recommends
- Call GET https://api.reworkd.dev/v1/outputs/{group_id} with an Authorization Bearer header and page through the cursor
Pros & Cons
Pros
- The whole chain sits in a single product: scanning, code generation, execution, validation and output
- Generating code instead of predicting values removes a whole class of LLM hallucination
- Maintenance is absorbed by the platform, with scrapers repairing themselves when a target site changes
- Fully managed infrastructure on every tier: proxies, headless browsers, queues and anti-bot handling
- A permanently free Hobby tier at $0/month with 10 concurrent browsers, API access, captcha solving and scheduled jobs
- A documented public REST API with code samples in six languages, and the Harambe SDK open-sourced on GitHub
- Unusually good debugging for a managed scraper: browser traces, console and network logs, and inline code diffs
Cons
- The product has been announced as sunset on February 6, 2025, and the application login no longer accepts any authentication method
- No DPA, no published sub-processor list and no security certification of any kind
- No documented way to exclude customer data from model training, and no hosting country or region disclosed
- A generic Termly privacy notice dated 9 June 2023, with its postal address block left empty
- Short retention by design: 90 days guaranteed for downloaded files, and three months at most for personal data after account closure
- Infrastructure usage (proxies, compute, anti-bot) is billed on top of the subscription, which makes the final bill hard to forecast
- English-only interface and documentation, no mobile app or browser extension, and a main-menu Docs link that returns a 404
Pricing & Plans
Reworkd offers a permanently free plan: the Hobby tier costs $0 per month and includes $10 of infrastructure credits. The lowest paid entry point is the Pro tier at USD 99 per month, which includes $49 of monthly credits, while the Enterprise tier is priced on request. Infrastructure consumption beyond the included credits is billed separately, at $0.125 per GB for standard proxies, $8 per GB for premium proxies, $0.10 per hour of compute and $5 per 1,000 anti-bot solves. No time-limited free trial, annual discount or promotional pricing is advertised.
- 10 concurrent browsers
- 30-day data retention
- API access
- captcha solving
- scheduled jobs
- fully managed solution and $10 of included credits
- 50 concurrent browsers
- 90-day data retention
- API access
- captcha solving
- scheduled jobs
- Slack support
- fully managed solution and $49/month of included credits
- custom concurrent browsers
- retention and credits
- API access
- captcha solving
- scheduled jobs
- Slack support
- a dedicated Slack channel and a fully managed solution
- custom on Enterprise
Data, GDPR & hosting
A consolidated view of how Reworkd handles your data.
GDPR overview
Reworkd never claims GDPR compliance anywhere on its site. Section 3 of the privacy notice explains that the GDPR and UK GDPR require it to state its legal bases, listed as consent, performance of a contract, legitimate interests, legal obligation and vital interests. Users in the EEA, the UK and Canada are granted access and copy, rectification, erasure, restriction of processing, portability and objection rights, with consent withdrawable at any time; the notice points to national supervisory authorities, the UK ICO and the Swiss data protection authority. Beyond that the file is thin: no Article 27 EU representative, no data protection officer, no dedicated privacy contact beyond support@reworkd.ai, no DPA and no sub-processor list. California CCPA and Virginia CDPA sections are present, with a 45-day response window extendable by 45 days. The notice dates from 9 June 2023 and its postal address block was left blank.
Who owns the data?
The terms of service leave the user responsible for the data they submit. Section 22, User Data, says Reworkd retains certain transmitted data to manage service performance and runs routine backups, while disclaiming any liability for loss or corruption and asking users to waive any right of action arising from it. On the personal side, the privacy notice covers names, email addresses and billing addresses, with no sensitive categories and no data received from third parties; payment details are held by Stripe. Scraped files sit in Reworkd's own S3 buckets and are retrieved through pre-signed URLs. Reworkd AI, Inc. states it has never sold personal data to third parties and commits not to.
Reuse rights
The material reviewed records no clause limiting how a customer may re-use the data their own jobs extract; the terms simply make the user solely responsible for what they transmit, and results are pulled out through bulk JSON or CSV exports and the public API without any permission step. Reworkd's own use is set out in the privacy notice: it processes information to provide, improve and administer the services, communicate with users, prevent fraud and comply with the law, and more specifically to create and authenticate accounts, answer requests and support tickets, request feedback and protect vital interests. Automatically collected data includes IP address, browser and device characteristics, operating system, language, referring URLs, approximate or precise location, usage logs and error reports, gathered through cookies, web beacons and pixels. Information may be shared in a merger, sale of assets, financing or acquisition, and with business partners for products, services or promotions. Signing up through a social account gives access to name, email, friend list and profile picture. Do-Not-Track browser signals are not honoured, and training models on customer data is not mentioned anywhere.
Data retention & training
Hosting summary
Reworkd names no hosting country and no hosting region anywhere on its site or in its documentation. What the documentation does say is that downloaded files are stored in Reworkd's own Amazon S3 buckets and served through pre-signed URLs, with asynchronous downloads handled by AWS Lambda and a dedicated queue. That establishes an AWS footprint without telling you which region it sits in. The publisher, Reworkd AI, Inc., is incorporated in California, which makes a US jurisdiction the reasonable expectation, though it is never stated as a hosting commitment. The marketing site itself resolves to 76.76.21.21, an anycast CDN node geolocated in the United States, but that describes where the website is served from, not where customer data lives. There is no DPA, no sub-processor list beyond Stripe as payment processor, and no data residency option on offer. European users who need a documented region or transfer basis will not find one here.
Things to keep in mind
Risks and trade-offs to weigh before adopting Reworkd.
- The site contradicts itself on the sunset date: the banner announces February 6, 2025, while the blog kept publishing product updates into May 2025
- The application's login screen no longer offers any authentication method, so continued access cannot be assumed
- The bill depends heavily on infrastructure consumption (proxies, compute, anti-bot solving), which is uncapped beyond the included credits
- Premium proxies cost 64 times more per GB than standard proxies, an easy way to run up a bill without noticing
- No data processing agreement is offered, so there is no contractual data protection commitment to fall back on
- The postal address block in the privacy notice is blank, and the footer address (550 15th St, CA 94103) differs from the legal address in the terms (2261 Market Street #5228, CA 94114)
- Scraping third-party sites puts the legal exposure on you: the terms make the user solely responsible for the data they transmit
Setup & Integrations
Technical difficulty
The marketing promise is no code, no maintenance, no worries, and there is genuinely nothing to install: the solution is fully managed and everything happens in the web application. In practice, getting started means creating a group, writing a structured data schema, adding source URLs and choosing a schedule, and the schema step assumes you can describe the shape of the data you want. Per-job settings such as proxy type and timeouts assume some understanding of the target site. Advanced use, with API keys, the Harambe SDK, browser traces and code diffs, is developer territory.
Deployment
Integrations
Supported languages
Behind Reworkd
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Frequently asked questions
Is Reworkd still available?
How does Reworkd actually extract data?
What happens when a target website changes?
Is there a free plan?
How much does the first paid tier cost?
Does Reworkd have an API, and in which formats can I get my data out?
Are linked files such as PDFs downloaded too?
How long is data kept?
Is there a mobile app?
What is the connection with AgentGPT?
Should you pick Reworkd?
Reworkd is a genuinely well-built piece of engineering that arrives with a hard caveat attached. The technical bet is sound: rather than asking a model to read values off a page, the agents write the extraction code, which sidesteps the hallucination problem that dogs LLM-based scraping. Around it sit self-healing scrapers, schema validation, deduplication, managed proxies and headless browsers, a documented REST API and an open-source SDK. The customer stories are concrete and numbered, with 2,500+ regulation sites for Axis and 350,000+ e-commerce products for a Gen Z fashion startup, and the backers read well, with Y Combinator, General Catalyst, AI Grant and Paul Graham behind $4M raised. Then there is the banner. Reworkd announces that the product is being sunset on February 6, 2025, the application's login screen no longer offers any authentication method, and yet the blog kept shipping product updates into May 2025. The site does not resolve its own contradiction, and we will not resolve it for it: anyone considering Reworkd today should treat availability as the first question to settle, not the last. The legal file does not help. A generic Termly privacy notice from June 2023 with an empty address block, no DPA, no published sub-processors, no hosting country, no Article 27 EU representative. Retention is short by design, at 90 days for downloaded files and three months for personal data after account closure. What survives regardless is Harambe, the Playwright-based SDK Reworkd open-sourced on GitHub. For most readers, that is now the practical takeaway.
- Choosing a selection results in a full page refresh.
- Opens in a new window.