What happens to a file after you upload it.
Valara reviews sensitive collateral, so it should be clear where your files go and who touches them. This page describes how documents move through the system today, backed by the code that runs it. Where we cannot yet make a firm commitment, we say so plainly.
How a file moves through Valara.
A single appraisal passes through four stages, from upload to a cited review in your dashboard.
- 01
Upload
You upload a PDF appraisal from the web app. The original file is stored in Vercel Blob storage and referenced by a per-document record in our database.
- 02
Parse and extract
The document is sent to Landing.ai (Agentic Document Extraction) to turn pages, tables, and photos into structured data. Parsed results are cached, keyed by a SHA-256 hash of the file content, so the same file is not reprocessed.
- 03
Review
The extracted data drives the review workflow. Analysis prompts run through the Vercel AI Gateway to an Anthropic Claude model (claude-haiku-4-5 by default). Every finding is tied back to the page and value it came from.
- 04
Results
Scores, findings, and citations are written to our PostgreSQL database and shown in your dashboard. Reviewer annotations you add are stored alongside the review.
Three stores, each with one job.
PostgreSQL (Neon / Vercel Postgres)
Accounts, document records, review scores and findings, reviewer annotations, and credit balances. Accessed through the Drizzle ORM.
Vercel Blob
The original uploaded PDF, referenced by a signed blob URL on the document record.
Vercel KV
Cached parse and extraction output, namespaced and keyed by the SHA-256 content hash of the file so repeat uploads reuse prior work.
Documents are scoped to the account that owns them.
Per-account ownership
Every read of a document checks that the requesting account owns it before returning data. Administrator accounts are a distinct role. When an account is deleted, its documents, annotations, and related records are removed with it through database cascade rules.
Authentication
The app uses Better Auth with email and password sign-in and server-managed sessions. Programmatic clients authenticate with bearer tokens. Sessions expire after seven days.
Who else can see your files.
These services process document content or account data on Valara’s behalf.
Vercel
Application hosting, plus Blob (file storage) and KV (cache).
Neon
Managed PostgreSQL that backs the application database.
Landing.ai
Agentic Document Extraction: parses uploaded PDFs into structured pages, tables, and fields.
Anthropic (via Vercel AI Gateway)
Runs the review analysis prompts. Requests are routed through the Vercel AI Gateway.
Resend
Transactional email (for example, signup and approval notifications).
Stripe
Payment processing for credit purchases.
The questions we have not answered here yet.
We would rather leave these blank than fill them with a promise we cannot yet back up. If one of them is a blocker for your team, reach out and we will walk you through where things stand.
Encryption specifics
We rely on our infrastructure providers for transport and storage encryption. We will publish the specific encryption posture (in transit and at rest) for each store here.
Data retention and deletion
Removing a document from your dashboard deletes its record from your account. We will document the full retention schedule and how the underlying file and cached extraction are handled after removal.
Model and subprocessor data use
We will document our subprocessors’ data-use and training terms, and our own policy on how uploaded content may or may not be used, before making any claim about it.
Related: how we evaluate review accuracy and what appraisal review means.
Have a security question we did not answer?
Talk to us before you upload, or try a sample review that uses no file of your own.
5 credits included, no card required.