One policy. Every place data moves.

Text API, playground, file studio.

POST text, get it back redacted — with detections, a risk profile and optionally a signed receipt. Upload PDFs, scans, spreadsheets and docs: Veil redacts in the original format (true black-box PDFs, image boxes, cell-level XLSX/CSV), lets reviewers keep or redact each item, and returns a same-format download plus receipt.

• Per-type strategies: redact, mask, hash, format-preserving encrypt, synthetic, keep
• Custom recognizers (regex) as first-class PII types
• OCR for scans with automatic language detection (13 packs in Docker)
• Date-of-birth vs ordinary dates, multilingual names & addresses (veil-pii v2)

curl -X POST https://veil.example/api/policies/1/redact?receipt=true \
  -H 'X-Veil-Api-Key: vk_…' -H 'Content-Type: application/json' \
  -d '{"text":"Priya Sharma, priya@example.com, card 4111 1111 1111 1111"}'

{ "redacted": "<PERSON>, <EMAIL>, card <CREDIT_CARD>",
  "detections": [{"type":"PERSON","start":0,"end":12,"score":0.93}, …],
  "receipt": { "id":"rcpt_…", "alg":"Ed25519", "manifest": { … } } }

Flow: payslips → HR system
  source   SFTP /incoming (or S3, IMAP, Drive, Dropbox, OneDrive, HTTP, portal)
  classify teachable type "Payslip" (pack) · confidence ≥ 0.8
  LOOKUP   employee by ID → HR API / JDBC / reference table
  CONSENT_CHECK purpose=hr_administration → ALLOW / REDACT / TOKENIZE / REVIEW / DROP
  REDACT   policy "hipaa-safe-harbor"  (same-format artifact)
  EXTRACT  → Kafka topic + webhook, payload.receipt attached
  residency EU · ENFORCE

Teach a document type in minutes. Ship a pipeline the same day.

Annotate a handful of examples, train and benchmark (leave-one-out), publish with rollback. Flows pull from storage, mail, drives and SaaS, extract fields, enrich and look up context from your systems, decide on consent, redact, and deliver — with review queues, retention, legal hold, SLA alerts, near-duplicate detection, AI-assisted auditing and reconciliation rules.

• 22 prebuilt packs (invoices, IDs, payslips, medical…) — 46 document types
• Sources: S3, SFTP/FTP, folders, IMAP, intake portals, Google Drive, Dropbox, OneDrive, HTTP APIs, 8 SaaS presets
• Sinks: Kafka, JDBC, webhooks, HTTP, drives; optional signed receipt on every delivery

Safe data for everyone who isn't production.

Discover and classify PII columns, then mask DB-to-DB with FPE and synthetic strategies that keep referential integrity, or cut a consistent subset across foreign keys. Every export can carry a re-identification risk gate: if k-anonymity over your quasi-identifiers falls below the floor, the job fails and the output is withdrawn.

• Postgres, MySQL, SQL Server, Oracle, Snowflake, Databricks via JDBC
• DSAR: find and export / erase a subject across sources
• Durable job queue: retries, lease recovery, checkpoints, idempotency, multi-replica

Clean in motion, clean at rest.

Stream redaction for Kafka / Redpanda clusters and bucket jobs for S3 / MinIO (CSV, NDJSON, text — line by line, checkpointed, resumable). Same policies, same vault, same metering.

Logs without the leaks.

Point your collector at Veil's OTLP logs endpoint or Splunk HEC; bodies and attributes are redacted and forwarded to Datadog, Loki, Splunk or any OTLP backend via forwarding profiles. Per-surface metrics land in Prometheus / Grafana.

exporters:
  otlphttp:
    endpoint: https://veil.example/v1/logs   # Veil redacts → forwards
    headers: { X-Veil-Api-Key: vk_… }

from openai import OpenAI
client = OpenAI(base_url="https://veil.example/ai/support-bot/v1", api_key="vgw_…")
# prompts: PII → vault tokens · blocked types → 403 · consent for the route purpose checked per call
# answers: detokenized for the caller's audience (X-Veil-Audience) · receipt in X-Veil-Receipt
# agents: POST /ai/support-bot/mcp — tool args & results get the same DLP, indirect injection blocked
# RAG:    POST /ai/support-bot/v1/ingest — documents cleaned + chunked before embedding

Let models reason over tokens, not people.

Wire-compatible with OpenAI and Anthropic SDKs. Redact or tokenize prompts, block listed types, rate-limit, flag or block prompt injection, enforce residency and consent per call, and return real values only to the audience that may see them. Agents get the same guarantees on every tool call; ingestion pipelines get redact-before-embed with a signed batch receipt.

Govern in Veil, enforce in the warehouse.

Turn a Veil policy and your classification into native masking policies — Snowflake, Databricks, BigQuery, Redshift, SQL Server, Oracle, Postgres (anonymizer) — generated, reviewed, applied.

Reversible, audited, audience-aware.

Deterministic tokens (same value → same token) across tables, streams and LLM hops. Reveal levels — full, partial, synthetic, none — per role and per audience; every reveal is audited. Consent-denied documents can be tokenized instead of dropped.