--- title: "Base64.ai Integration" description: "Extract structured data from documents with Base64.ai. Agents scan IDs, invoices, receipts, and forms into labeled fields with confidence scores, detect and match faces and signatures for identity verification, and redact sensitive fields before storage." url: https://flowrunner.ai/integrations/base64-ai date_modified: 2026-08-04T10:04:16-07:00 --- # Base64.ai [Document Processing](https://flowrunner.ai/integrations/category/documents-forms) Extract structured data from documents with Base64.ai. Agents scan IDs, invoices, receipts, and forms into labeled fields with confidence scores, detect and match faces and signatures for identity verification, and redact sensitive fields before storage. [Verified](https://flowrunner.ai/integrations/verified "What does verified mean?") · 11 actions · API key · available [Base64.ai website](https://base64.ai/) · [Platform Documentation](https://base64.ai) · Capability data verified 2026-07-27 1. A new customer submits an ID photo and a selfie through the onboarding portal 2. Scan Document extracts the ID's labeled fields, name, date of birth, document number, with per-field confidence scores 3. Recognize Face (Match) compares the ID's photo against the selfie and returns a similarity score 4. High-confidence extractions with strong face-match scores pass automatically; the workflow sorts everything else into a review queue 5. Sensitive fields are redacted before the document image is stored 6. Borderline cases post to the verification channel with the extracted fields, both images, and the scores 7. The compliance reviewer examines the borderline match and makes the accept-or-reject call, because a verification verdict lands on a real person ## What This Integration Enables Base64.ai answers the question that decides whether document AI belongs in production: not "what does this document say?" but "how sure are you?" Every extracted field arrives with a confidence score, every face and signature match arrives as a similarity score, and those numbers are precisely what a FlowRunner workflow needs to split traffic: the confident cases flow through untouched, and the uncertain ones route to a person. That split is the entire architecture of trustworthy document automation, and this connector ships with it built in. - Extract labeled fields, text, tables, images, and signatures from IDs, passports, invoices, receipts, and forms in a single stateless call - Restrict recognition to specific model types and redact sensitive fields before storage - Process large files and high-throughput batches asynchronously with job polling - Detect and match faces for identity verification, including ID-to-selfie comparison - Detect and match handwritten signatures against known specimens - Reuse configured extraction flows and review stored results with pagination - Monitor plan, quota, and remaining usage from inside the workflow ### Without FlowRunner **Document data is retyped by hand**: Invoices, IDs, and forms queue for manual entry, and every keystroke is a chance to be wrong **Every verification needs eyes**: A person compares every ID photo to every selfie, including the 95 percent that are obviously fine **Sensitive documents are stored raw**: Full ID images with names, birthdates, and document numbers sit in storage unredacted ### With FlowRunner **Fields extract with confidence attached**: Documents become labeled data, and every field carries a score that says how much to trust it **Humans review the borderline, not the pile**: Clear passes and clear fails resolve automatically; judgment is spent where scores are ambiguous **Redaction happens at ingestion**: Sensitive fields are removed at scan time, so what reaches storage is what should reach storage ## Use Case Scenarios ### The invoice queue that empties itself Vendor invoices land in [Dropbox](https://flowrunner.ai/integrations/dropbox-service). Each new file triggers Scan Document restricted to the invoice model, and the labeled fields, vendor, date, line items, totals, flow into [Google Sheets](https://flowrunner.ai/integrations/google-sheets) and the accounting system. Fields above the confidence threshold post straight through; any invoice with a low-confidence total or an unrecognized layout posts to [Slack](https://flowrunner.ai/integrations/slack) with the cropped field image for a human to confirm. The AP team stops typing and starts adjudicating, and only the documents that earned adjudication. ### Identity verification with a human on the margin Onboarding runs the two-check gauntlet: Scan Document on the ID, Recognize Face (Match) between the ID photo and the selfie. Scores above the pass threshold verify automatically, and scores below the fail floor decline with a resubmission prompt. The band between routes to the compliance queue with fields, images, and scores assembled. Reviewers see a few dozen genuinely ambiguous cases a day instead of every applicant, and every automated decision leaves the scores in the audit log as its stated reason. ### Signed means verified, not just present Executed agreements flow through Detect Signatures to confirm every required signature block is actually signed, with unsigned documents bounced back to the sender automatically. For high-value agreements, Verify Signature (Match) compares the signature against the counterparty's specimen from the vault, and low similarity scores escalate to legal with both crops side by side. Batch backfills of the contract archive run through Start Async Scan and Get Async Scan Result, so a decade of paper becomes queryable data without blocking anything. ## Human-in-Loop Highlight Recognize Face (Match) returns a similarity score, and somewhere a workflow turns that number into a verdict on a human being: verified or rejected. Auto-rejecting a legitimate customer because the lighting in their selfie dragged the score down is a customer-facing failure you cannot fully take back; auto-approving a fraudulent match is worse. Base64.ai's own design points at the answer, because per-field confidence and similarity scores exist to be routed on. FlowRunner makes the routing explicit: the workflow owner sets the pass threshold and the fail floor, and everything between goes to a person with the evidence assembled, both images, the extracted fields, the scores. The reviewer decides the margin. The system's honesty about its own uncertainty is what makes the automation safe to run at volume. Agent processes routinely Detects exception requiring judgment Clear match Continues automatically Ambiguous Routes to human via preferred channel Human decides Agent resumes with decision ## Agent Capabilities 11 actions ### Document Extraction 3 - **Scan Document** Extracts structured data from a document or image in a single stateless call, automatically recognizing the model type: IDs, passports, driver's licenses, invoices, receipts, forms. Returns labeled fields, text, tables, images, signatures, and per-field confidence scores, with optional model restriction, field redaction, and advanced settings. - **Start Async Scan** Submits a document for asynchronous processing and returns a job UUID immediately. The entry point for large files and high-throughput batches. - **Get Async Scan Result** Retrieves the result of an asynchronous scan by job UUID, returning the extraction when complete or a status while in progress. Polled until the result is available. ### Face Recognition 2 - **Detect Faces** Detects and analyzes human faces in a document or image, returning each face's bounding location, cropped image, and confidence. Used to extract the photo from an ID or confirm a selfie contains a face. - **Recognize Face (Match)** Compares a face in a reference document against a face in a query image and returns a similarity score for identity verification. The score this page's human gate routes on. ### Signature 2 - **Detect Signatures** Detects handwritten signatures in a document, returning each signature's location, cropped image, and confidence. The check that confirms agreements are actually signed. - **Verify Signature (Match)** Compares a signature in a reference document against a query document and returns a similarity score indicating whether the same person produced both. Used to verify signatures against known specimens. ### Flows 1 - **List Flows** Lists the configured extraction flows available to the account, each bundling model selection, validation, and post-processing rules, with IDs for use in scans and result queries. ### Results 2 - **Get Flow Results** Retrieves stored scan results for a flow, ordered by time, with text filtering and cursor-based pagination in both directions. The review path over previously processed documents. - **Get Result by UUID** Retrieves a single stored scan result by UUID, including the recognized model, fields, and features. Used to re-fetch a specific extraction record for audits and reprocessing. ### Account 1 - **Get Account** Retrieves the authenticated account's profile, subscription plan, and remaining usage or credit information. The quota check before large batches, and the credentials sanity test. ## Frequently Asked Questions ### What can FlowRunner do with Base64.ai? FlowRunner agents can run Scan Document, Start Async Scan, and Get Async Scan Result in Base64.ai, plus 8 more actions. ### Does connecting Base64.ai to FlowRunner require OAuth? No. Base64.ai connects to FlowRunner with an API key, no OAuth flow required. ### Can Base64.ai trigger a FlowRunner workflow automatically? Base64.ai doesn't currently expose triggers in FlowRunner. It connects as an action step inside workflows started by another trigger. **Work at Base64.ai?** This integration exposes Base64.ai to AI agents on every FlowRunner plan, including through MCP, at no cost to you. [See what FlowRunner offers integration partners](https://flowrunner.ai/integrations/partners), including how to keep this page current. --- Markdown version of https://flowrunner.ai/integrations/base64-ai. Site index: https://flowrunner.ai/llms.txt