Cradl AI
Cradl AI is a no-code platform for building document-processing agents. It extracts structured data from invoices, contracts and PDFs, checks it against validation rules, sends uncertain results to human reviewers, then delivers clean JSON to n8n, Zapier or Power Automate.
What is Cradl AI?
Cradl AI is a cloud platform for building AI agents that read business documents and hand back structured data. It is published by Lucidtech AS, a Norwegian company registered in Oslo under organisation number 918 345 787, and it targets a specific gap: teams that already automate with n8n, Zapier, Make or Microsoft Power Automate but cannot get document steps reliable enough for production.
The product follows four stages. An agent extracts the fields you define from a document, validation rules check the result, anything uncertain is routed to a human review interface, and the confirmed output is exported as JSON. Each prediction carries a confidence score from 0 to 100, and you set the threshold at which a document stops being processed automatically and waits for a person. Reviewers work in a purpose-built interface, can be invited with a restricted Reviewer role, and every correction they make feeds back into the agent so accuracy improves over time.
Document coverage is broad: digital and scanned PDFs, invoices and receipts, contracts and forms, reports and statements, images containing text, documents with tables and line items, and handwritten pages. Format validators cover IBAN, BIC/SWIFT, VAT and tax identifiers, passport and phone numbers, emails, freight container IDs and licence plates, and a match-and-map validator aligns extracted values with an existing reference list.
Two delivery modes exist and only two: a web application and a REST API at api.cradl.ai/v1 with OAuth2 bearer authentication. Native connectors ship for n8n, Zapier, Make and Power Automate, alongside webhook, mailhook, Excel, Google Drive, OneDrive and Google Sheets. A dedicated Cradl AI node was published in the n8n library in March 2026.
Under the hood the platform runs Qwen 3.vl as the default model for new agents, with automatic fallback to a secondary model during outages, and a second-opinion option that cross-checks predictions against another model. Documents are stored on Amazon Web Services in Ireland. Named customers include Equinor, DNB, SR-Bank and Posten Bring, and the published service level agreement commits to 99.6% monthly uptime.
What it does
- Extract fields, tables and line items from invoices, receipts, contracts, forms, statements and scanned or handwritten documents
- Score every prediction from 0 to 100 and send anything below your threshold to human review
- Apply validation rules and format checks for IBAN, BIC/SWIFT, VAT and tax IDs, passports, phone numbers, emails, freight container IDs and licence plates
- Map extracted values onto an existing reference list so they land correctly in an ERP, CRM or database
- Push clean JSON straight into n8n, Zapier, Make, Power Automate, a webhook or a spreadsheet
- Cross-check a prediction against a second model before it moves downstream
- Track accuracy, exceptions, throughput and reviewer productivity from a monitoring dashboard
When to use Cradl AI / When not to
A quick filter to help you decide if Cradl AI is the right fit.
When to use Cradl AI
- Automation specialists and citizen developers already building flows in n8n, Zapier, Make or Microsoft Power Automate
- Finance and accounts-payable teams handling recurring invoices, receipts and statements
- Operations and shared-service teams moving a document process from manual keying to production automation
- Procurement, logistics and contract teams that need fields pulled reliably from order forms, price lists and agreements
- Regulated organisations in banking or insurance that require EU hosting, a signed DPA and a full audit trail
When not to use Cradl AI
- Teams that want an open-source engine they can self-host and drive entirely from code
- Anyone looking for plain OCR that only turns an image into raw text
- Users after a general-purpose assistant for chatting, drafting or summarising
- Mobile-first workflows, since the product ships only as a web application and a REST API
- Very high page volumes on the entry tiers, where quotas stop at 50 pages a month on Free and 500 on Builder
How to use Cradl AI
A typical end-to-end flow, from setup to results.
- Create an account from the Start for free button on cradl.ai
- Copy a ready-made agent template that matches your document type, or start from the Custom option
- During onboarding, declare how the agent will feed your workflow, such as a spreadsheet, a database or a third-party app
- Adjust the fields the agent should capture and the validation rules it should apply
- Import a test document from the Runs view, or use the sample document supplied with the template
- Open the review interface from the Runs table and check the predictions field by field
- Correct anything wrong, then click Confirm to release the data; every edit trains the agent
- Set the confidence threshold that decides which documents go straight through and which wait for review
- Configure the export integration, since nothing is exported automatically until you do
- For advanced use, request a token from auth.cradl.ai/oauth2/token and call the REST API at api.cradl.ai/v1
Pros & Cons
Pros
- Human review, confidence thresholds and exception handling are built into the product rather than assembled by hand
- Native connectors for the four main automation platforms remove the need for custom integration code
- Documents are hosted on AWS in Ireland, with processing kept mainly inside the EU/EEA and a named sub-processor
- The full legal stack is public: terms, DPA, EULA, SLA, description of service and privacy policy
- A genuinely permanent free plan allows evaluation without a time limit or a commitment
- A published SLA commits to 99.6% monthly uptime with service credits
- Every extraction and human correction is logged, giving a usable audit trail for regulated processes
Cons
- Page quotas are tight on the lower tiers, at 50 a month on Free and 500 on Builder
- Data retention controls are absent from Free and Builder, while role-based access control and SSO are Enterprise-only
- The DPA and the description of service disagree about whether posted documents may be used to train Cradl AI's models
- Documents may be kept for up to five years on the vendor's systems under the DPA
- No security certification such as SOC 2 or ISO 27001 is named anywhere on the site
- There is no mobile app, no self-hosted edition and no open-source version
- No list of supported interface or document languages is published
Pricing & Plans
A permanent free plan is available, described as free forever and covering up to 50 pages per month with one agent and one seat. The cheapest paid entry point is the Builder plan at USD 40 per month when billed annually, or USD 50 per month on the monthly option. Pro follows at USD 240 per month annually, or USD 300 monthly, and the Enterprise plan is quoted on request.
- up to 50 pages per month
- 1 agent
- 1 seat
- all integrations
- chat support
- table extraction and workflow builder
- up to 500 pages per month
- 2 agents
- 2 seats
- full and priority support
- up to 5
- 000 pages per month
- 5 agents
- 5 seats
- priority support
- service level agreement and data retention controls
- custom volumes
- volume discounts
- single sign-on
- role-based access control
- dedicated account manager and custom billing
Data, GDPR & hosting
A consolidated view of how Cradl AI handles your data.
GDPR overview
GDPR handling is documented rather than merely claimed. Lucidtech AS is the controller for its own site data and a processor for customer documents, under a published data processing agreement. The privacy policy cites Article 6(1)(b) and 6(1)(c) plus legitimate interest as legal bases, lists access, rectification, erasure, restriction and portability rights with a normal 30-day response, and names Datatilsynet, the Norwegian supervisory authority, for complaints. Processing takes place mainly inside the EU/EEA, with standard contractual clauses for any transfer outside it. Amazon Web Services EMEA SARL is named as the approved sub-processor, breaches are notified within 24 hours, and sub-processor changes carry three months' notice. Contact is privacy@cradl.ai. No security certification such as SOC 2 or ISO 27001 is named anywhere on the site.
Who owns the data?
Everything you upload stays yours. The terms of service state that all content posted to or processed by the platform remains the sole property and responsibility of the licensee, and that Cradl AI carries no liability for it. On the GDPR side the customer, or the customer's own client, is the data controller and Lucidtech AS acts only as a processor under its published data processing agreement. The vendor keeps ownership of the other half: the service itself, its APIs, workflows, machine-learning models and trained model weights are its intellectual property, and using the product transfers none of those rights to you.
Reuse rights
Customers may reuse the extracted data freely in their own systems without asking permission: the output is returned as structured JSON and the content remains the licensee's property. Deletion is on demand at any time through the API, and on termination Cradl AI undertakes to erase all licensee and end-user data. The vendor's own use is narrower on paper. The data processing agreement restricts processing to delivering the service, allows model training only at the controller's explicit request, and forbids using customer data to train or improve underlying base models. The description of service is looser, saying documents posted to the service may be used to train Cradl AI's models through fine-tuning or in-context learning, with automatic anomaly checks and occasional random inspection by employees. Those two statements do not fully agree and are worth clarifying before signing.
Data retention & training
Hosting summary
Documents are stored on Cradl AI servers hosted by Amazon Web Services in Ireland, according to the published description of service. The data processing agreement names Amazon Web Services EMEA SARL as the approved sub-processor and states that personal data is not processed outside the EU/EEA without the controller's prior written approval. The privacy policy confirms that processing takes place mainly within the EU/EEA and that any transfer beyond it relies on the European Commission's standard contractual clauses. Jurisdiction is Norwegian: Lucidtech AS is based in Oslo, the terms are governed by Norwegian law, and disputes go to Oslo District Court. Two caveats are worth noting. The site's public IP resolves to an anycast Amazon node in Amsterdam, which is a content delivery point and not a storage location. Separately, the site's Google Analytics usage sends analytics data to Google servers in the United States, which is distinct from where customer documents are held.
Things to keep in mind
Risks and trade-offs to weigh before adopting Cradl AI.
- The vendor states that predictions are suggestions and that deciding whether to trust them remains the customer's responsibility, so accountability for a wrong figure stays with you
- Automated extraction encourages people to stop reading documents; if reviewers rubber-stamp what the confidence score approves, the safety net quietly disappears
- Confidence thresholds set too permissively to save review time convert directly into bad data reaching finance or ERP systems
- Documents may be stored for up to five years, and retention controls are unavailable on the Free and Builder plans
- Cradl AI employees may carry out random inspection of collected data as part of the training process described in the description of service
- Liability is capped at the annual contract value and the service is delivered as is, which is thin cover for a business-critical document flow
- Google Analytics is active on the site with data transmitted to Google servers in the United States
Setup & Integrations
Technical difficulty
Low for the intended audience. Onboarding is four steps: pick an agent template, declare how it will feed your workflow, import a test document, then review and confirm. No code is required, and an Equinor reviewer quoted on the site reports going from idea to production in under a day. Two things still need attention: nothing is exported until you configure an export integration, and tuning confidence thresholds takes a few real documents to get right. Technical teams have a REST API with OAuth2, webhooks and custom code available if they want them.
Deployment
Integrations
Behind Cradl AI
Fundraising
Social
Resources
All the official URLs gathered for verification and reference.
Alternatives
Tools that compete with or complement Cradl AI.
Frequently asked questions
What does Cradl AI actually do?
Do I need to know how to code?
Which tools does it connect to?
What kinds of documents can it handle?
How does it avoid acting on a wrong extraction?
Where is my data hosted?
How long are my documents kept?
Is my data used to train models?
Is there a free plan?
Is there an uptime commitment?
Should you pick Cradl AI?
Cradl AI knows exactly what it is: not a general assistant and not another OCR engine, but the document step that automation platforms handle badly. The pitch holds together. Confidence scores, adjustable thresholds, a human review interface and a validation library are all built in, which is precisely the machinery teams otherwise stitch together themselves and then maintain. Native connectors for n8n, Zapier, Make and Power Automate mean the tool slots into an existing flow rather than replacing it.
The governance story is unusually complete for a company of this size. Terms, DPA, EULA, SLA, description of service and privacy policy are all public, documents sit on AWS in Ireland, Amazon Web Services EMEA SARL is named as sub-processor, and the SLA commits to 99.6% monthly uptime with service credits. Named customers such as Equinor, DNB, SR-Bank and Posten Bring lend weight to the enterprise claims.
Three reservations deserve attention before committing. The DPA and the description of service do not tell the same story about training on customer documents, and that contradiction is on the vendor's own site. Documents may be retained for up to five years, while retention controls only appear on Pro and Enterprise. And no security certification is named anywhere, only a general reference to leading information security standards, which will matter to procurement teams that ask for SOC 2 or ISO 27001.
Page quotas are the other practical limit: 50 a month on Free and 500 on Builder run out quickly at real volumes. The free-forever plan makes testing genuinely cheap, so the sensible route is to run your own documents through it and judge the extraction quality before scaling.
- Choosing a selection results in a full page refresh.
- Opens in a new window.