All apps

Redactor

x402

Strip the PII out before the model sees it. Put it back after.

Redactor screenshotHover to scroll

Strip the personal and secret data out of text before it reaches a model, a log or a third party — then put it back after. Deterministic rules and checksums, no model in the path, and nothing you send is stored.

  • Nine entity types, each decidable from the text itself: email, phone, payment card (Luhn plus a known network prefix), IBAN (mod-97), US SSN, IP address, vendor-prefixed API key, JWT (the header has to really decode to JSON), and PEM private key blocks
  • Person names, postal addresses and dates of birth are deliberately NOT detected, and the app says so in the hero, in every API response and in the MCP capability tool. A redactor that silently misses a category is worse than none, because the caller stops looking
  • Per-detector precision and recall are published on the homepage, measured against a labelled corpus, with a test that recomputes them on every build so the marketing figure cannot drift from the code
  • Every response carries an unsure count: things that looked like they might matter and did not clear the bar (a Luhn-valid number with no card prefix, ten bare digits in dialling shape, a high-entropy string beside a secret-shaped label). They are reported with offsets and left in the text, never quietly removed
  • tokenize returns a stable mapping you can reverse; mask is irreversible by design and returns nothing that could undo it
  • Nothing is retained — not your text, not the redacted output, not the mapping, and not in the logs. Only counts are stored, and a test reads the database migration and fails the build if a column appears that could hold any of it
  • 1 credit per 10,000 characters, rounded up. Rehydrating bills the same way because it runs; reading a past run is free
  • Agent-native from day one: rdc_live_ REST keys, an MCP server (works with the Ounie AI Team via ?api_key=), keyless x402 pay-per-call in USDC on Base, and the fleet master key

For agents

Works with Claude, ChatGPT, and any MCP client. One server, every app: paste the gateway config and the agent can discover and run Redactor — and the rest of the store — through ounie_list_apps → ounie_describe_app → ounie_invoke_app.

{
  "mcpServers": {
    "ounie": {
      "url": "https://mcp.ounie.com"
    }
  }
}

Scoped endpoint — just this app. Point a client here when the agent should only ever see Redactor's tools:

{
  "mcpServers": {
    "redact": {
      "url": "https://redact.ounie.com/api/mcp"
    }
  }
}

To spend your Ounie credits on the scoped endpoint, append your app key to the URL as ?api_key=… — the header-less form clients like the Ounie AI Team need.

Credits used

Redact or restore, per 10,000 characters1 credit per call
Redactor — an Ounie app for AI agents · Ounie