Documentation

Build with Zanitize

A REST API for anonymize, rehydrate and detect; an OpenAI- and Anthropic-compatible gateway; and an MCP proxy for agents.

Quickstart

Create an API key in the console (scopes anonymize and rehydrate), then anonymize text. Keep the returned session_id: it is the conversation whose mappings live in the vault.

bash
curl https://api.zanitize.nada.software/v1/anonymize \
  -H "X-API-Key: $ZANITIZE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Email Jane Doe at jane@acme.com", "surrogate_mode": "realistic"}'

Send the anonymized text to any model. Then restore the originals in the model's answer with the same session:

bash
curl https://api.zanitize.nada.software/v1/rehydrate \
  -H "X-API-Key: $ZANITIZE_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"text": "Sure, I will email Maria Lopez today.", "session_id": "ses_123"}'

Authentication

Send your key as X-API-Key: zk_live_... or Authorization: Bearer zk_live_.... Keys belong to one tenant and carry scopes such as anonymize, rehydrate, detect, proxy, mcp and files. Full keys are shown once; rotate them from the console.

Context headers

  • X-Zanitize-Session: conversation id (vault scope conversation)
  • X-Zanitize-Scope: request, conversation or tenant
  • X-Zanitize-Policy: policy id to apply
  • X-Zanitize-Shadow: true: evaluate the policy without changing content
  • X-Zanitize-Surrogates: realistic (default), token or universe (Fun mode)
  • X-Zanitize-Locale, X-Zanitize-App, X-Zanitize-Universe

LLM gateway

Change your SDK's base URL. Zanitize anonymizes messages, system prompts and tool results, calls the provider, and rehydrates text and tool-call arguments in the response, including streams.

OpenAI (Python)

python
from openai import OpenAI

client = OpenAI(
    base_url="https://api.zanitize.nada.software/v1/proxy/openai/v1",
    default_headers={"X-API-Key": os.environ["ZANITIZE_API_KEY"],
                     "X-Zanitize-Session": "ticket-4812"},
)
resp = client.chat.completions.create(
    model="gpt-4o",
    messages=[{"role": "user", "content": "Summarize Jane Doe's ticket"}],
    stream=True,
)

Anthropic (Python)

python
import anthropic

client = anthropic.Anthropic(
    base_url="https://api.zanitize.nada.software/v1/proxy/anthropic",
    default_headers={"X-API-Key": os.environ["ZANITIZE_API_KEY"]},
)
msg = client.messages.create(
    model="claude-sonnet-4-5", max_tokens=1024,
    messages=[{"role": "user", "content": "Draft a reply to Jane Doe"}],
)

Provider credentials come from your tenant configuration or are passed through from the request (Authorization for OpenAI, x-api-key for Anthropic).

MCP proxy

Register an upstream MCP server in the console, then point your MCP client at https://api.zanitize.nada.software/v1/mcp/<server_id> (Streamable HTTP). On tools/call Zanitize rehydrates arguments going to the tool and anonymizes the results going back to the model.

json
{
  "mcpServers": {
    "crm-via-zanitize": {
      "type": "http",
      "url": "https://api.zanitize.nada.software/v1/mcp/crm",
      "headers": { "Authorization": "Bearer zk_live_..." }
    }
  }
}

The console's MCP page generates this configuration for Claude Desktop, Cursor and other clients.

Policies

Policies are versioned YAML. Rules are evaluated in order and the first match wins. Criteria include entity_types, JSONPath fields, tools globs and tool_direction, apps, routes, destination_providers and destination_models.

Actions: allow, redact, mask, pseudonymize, tokenize, generalize, date_shift, hash, block. A block returns 422 policy_blocked unless the policy is in shadow mode.

yaml
id: default
default_action: pseudonymize
scope: conversation
ttl_seconds: 86400
surrogate_mode: realistic
shadow: false
allow_list: ["Acme Support"]
rules:
  - id: mask-cards
    entity_types: [CREDIT_CARD]
    action: mask
    params: {keep_last: 4}
  - id: block-ssn-external
    entity_types: [SSN]
    destination_providers: ["openai"]
    action: block

Files

POST /v1/files/anonymize accepts multipart/form-data with .txt, .md, .csv, .json, .pdf, .docx, .png or .jpg. PDF/DOCX and OCR depend on optional server extras; without them the API answers 501 not_configured.

bash
curl https://api.zanitize.nada.software/v1/files/anonymize \
  -H "X-API-Key: $ZANITIZE_API_KEY" \
  -F file=@contract.pdf -F output=text

Vault and retention

Mappings are encrypted per tenant. Inspect a session's counts with GET /v1/vault/sessions/{id}, crypto-shred it with DELETE, or shred the whole tenant with POST /v1/vault/shred. Retention (TTL) is set per policy and in tenant settings.

Errors

Every error has the same body:

json
{"error": {"code": "policy_blocked", "message": "SSN may not be sent to openai", "request_id": "req_..."}}
  • 401 invalid credentials, 403 scope or role denied
  • 422 policy_blocked a policy blocked the content
  • 429 rate_limited with Retry-After
  • 501 not_configured optional integration missing, 502 upstream_error from a provider