Klarefi is the complete platform for building and deploying AI intake agents. An intake agent rebuilds your form, talks to your applicants, extracts each fact with a source quote and page number, verifies it against your systems, and chases what is missing. Your team opens a complete, cited case file and handles the decision.

A processed intake is the first time Klarefi hands the completed intake to the operator queue. By then documents are received, facts are extracted with citations, gaps have been chased where possible, verifications have run, and a case file is assembled for your team to review. A single intake can contain multiple documents.

A fact is a specific piece of information your workflow needs from a document: a name, a date, a policy number, a coverage amount. You define facts when you build the agent: the label, a description, and the expected type. When documents come in, the agent locates each fact in the source text, extracts it, and attaches the exact quote and page number as evidence.

Self-serve is pay as you go per processed intake: one operator handoff, one usage unit, even when an intake includes many documents. Branding is an optional monthly add-on. Pilots and enterprise contracts are sales-led when you need SSO, custom terms, data residency, SLAs, committed volume, or procurement.

Operations teams in regulated industries: insurance claims, loan applications, KYC onboarding, medical intake, permit requests. If your team spends hours reading documents, cross-checking details, and chasing missing information before making a decision, an agent does that work for you.

Pilots typically run 2 to 4 weeks on your actual documents. You get measurable accuracy and coverage numbers by the end, and rollout follows when the numbers hold.

We assess fit, usually within one business day. If your use case aligns, we schedule a short call, walk through the product on your workflow, and define target metrics. We build your agent and run it on your actual documents. You decide on the measured results.

No. The agent sits on top of your core systems and runs the operational layer in front of them, starting with intake. Documents come in from your existing channels. Structured outputs push to your backoffice, DMS, or CRM via API and webhooks.

Your team reviews and selects from candidates. The agent assists, it does not decide autonomously. Corrections are recorded as tenant-scoped feedback and can inform reviewed workflow improvements; they do not train shared models.

EU data residency by default. Applicant documents and case content are not used to train shared models. Privacy-minimized operational telemetry may be used to operate, evaluate, and improve Klarefi. Deployment-specific residency and retention commitments are set out in your agreement.

Any unstructured document: PDFs, scanned pages, emails, photos, attachments. One case might have 5 to 30 documents depending on complexity. Klarefi structures them all.

No. The agent works on day one without training data. Workflows are configured through Form, Fact, and Verify, not trained. Operator corrections are retained as tenant-scoped feedback that can be used to evaluate and review improvements; historical labeled data is not required to start.

Yes, by design. An applicant can hand the session link to their own AI assistant (ChatGPT, Claude, or any agent runtime), and it can read the session state, fill the form, answer follow-up questions, and upload documents through the same API the hosted page uses. Your own backend or agents can do the same to prefill what your systems already know, over REST, the TypeScript SDK, or MCP. Agents can only complete the intake. Approval and every decision stay with your team.

That is exactly why we start with a pilot. We test Klarefi on your actual documents and define the target metrics together. You decide on the measured results, not on a demo.