Description

n8n-nodes-brikko


> Часть Brikko Privacy Ecosystem — open-source инфраструктура маскинга персональных данных перед AI для русского рынка.

> Privacy-aware n8n nodes by Brikko. Mask Russian and
> English PII before any data leaves your workflow for an LLM. Built for
> teams that need 152-ФЗ compliance.
>
> Узлы n8n с защитой персональных данных: маскируем PII перед отправкой
> в LLM. Для бизнеса РФ, работающего по 152-ФЗ.

This is an n8n community node — install через Settings → Community Nodes в self-hosted n8n.

—

What you get / Что внутри

Four community nodes:

| Node | What it does | Что делает |
|—|—|—|
| Brikko Anonymize | Replaces PII (name, email, phone, INN, SNILS, card, etc.) with placeholders like 001>. Returns the masked text plus a mappingid. | Заменяет ПДн на плейсхолдеры. Возвращает маскированный текст и mapping_id. |
| Brikko Restore | Inverse of Anonymize. Pass it the masked text + mappingid to get the original back. | Обратная операция: по mappingid восстанавливает оригинал. |
| Brikko Chat | One-node pipeline: anonymize → call a Brikko Gateway LLM (Claude / GPT / YandexGPT / GigaChat / DeepSeek / Gemini) → restore. | Полный конвейер в одном узле: маскирование → LLM → восстановление. |
| Brikko Detect PII | Scans text without modifying it. Returns categories, counts, and (optionally) sample values. Use it for audits and compliance gates. | Только обнаружение, без изменения текста. Категории, счётчики, примеры. |

Three operation modes / Три режима работы

You pick one per node:

1. Auto (default) — try the local Brikko Studio
sidecar at localhost:8403. If that’s not running, fall back to the
hosted Brikko Gateway at api.brikko.ru. If neither is configured,
degrade to a built-in regex masker.
2. Local Studio — only the sidecar. PII never leaves the machine.
3. Gateway — hosted only. No local install required.
4. Regex — pure offline regex with checksum validators for INN, SNILS,
and bank cards. Lower fidelity than NER but zero network calls.

—

Install / Установка

Via the n8n UI / Через интерфейс n8n

1. Settings → Community Nodes
2. Enter n8n-nodes-brikko and click Install
3. Reload the n8n editor.

Via npm (self-hosted) / Через npm

cd ~/.n8n
npm install n8n-nodes-brikko
restart n8n

Local development / Локальная разработка

git clone https://github.com/brikkoAI/n8n-nodes-brikko
cd n8n-nodes-brikko
pnpm install
pnpm run build
link into a local n8n instance
mkdir -p ~/.n8n/custom
ln -s "$PWD" ~/.n8n/custom/n8n-nodes-brikko

—

Credentials / Учётные данные

You configure these once under Credentials in the n8n UI.

Brikko API (for hosted Gateway / Шлюз)

| Field | Default | Notes |
|—|—|—|
| API Key | — | Get one at brikko.ru |
| Base URL | https://api.brikko.ru | Override only for staging |

Brikko Local Studio API (for the on-prem sidecar)

| Field | Default | Notes |
|—|—|—|
| Studio URL | http://localhost:8403 | Where your Brikko Studio sidecar listens |
| Workspace ID | default | Scopes the per-workspace mapping store |

You can configure both — auto mode will use whichever is reachable.

—

Example workflows / Примеры сценариев

1. Bitrix24 lead → privacy-aware LLM reply / Брикко-чат для Битрикс24

The most common Russian usecase: a new lead lands in Bitrix24, you want an
LLM to draft a personalised reply, but you can’t ship the customer’s name,
phone, and email to OpenAI in the clear.

Bitrix24 Trigger (new lead)
        ↓
Brikko Chat
  prompt:   "Напиши приветствие для лида: {{ $json.NAME }} {{ $json.PHONE }}, интерес: {{ $json.COMMENTS }}"
  model:    claude-sonnet-4-6
  policy:   balanced
        ↓
Bitrix24: Create Comment on Lead
  text: {{ $json.brikko_chat.response }}

What happens under the hood:
1. Brikko masks the prompt → Напиши приветствие для лида: 001> 001>, интерес: ...
2. Sends the masked version to Claude via Brikko Gateway.
3. Restores 001>/001> in Claude’s response before writing
back to Bitrix24.
4. The audit log on the Anonymizer side records exactly which categories
were touched, with no plaintext PII.

2. CSV upload → audit which rows leak PII / Аудит CSV

Marketing exports a CSV and asks “which rows contain personal data?”

Read Binary File (input.csv)
        ↓
Spreadsheet File (parse rows)
        ↓
Brikko Detect PII
  text:       {{ JSON.stringify($json) }}
  categories: [INN, SNILS, CARD, EMAIL, PHONE]
  includeSamples: false
        ↓
IF $json.brikkopii.totalcount > 0
        ↓                ↓
   write to flagged.csv  pass through

Detect-only mode never rewrites your data — it just tells you what’s
there. Sample values default to placeholders (not raw PII) when you call
through Studio or Gateway.

3. Manual three-node control / Ручной 3-узловый pipeline

When you want to use a third-party LLM provider directly (e.g. you already
have an OpenAI account) but still want the masking layer:

Slack Trigger (new message in #support)
        ↓
Brikko Anonymize
  text: {{ $json.text }}
  policy: strict
        ↓
OpenAI: Chat
  prompt: {{ $json.brikko.masked_text }}
  model: gpt-4o
        ↓
Brikko Restore
  text:       {{ $json.choices[0].message.content }}
  mappingId:  {{ $('Brikko Anonymize').item.json.brikko.mapping_id }}
        ↓
Slack: Send Message
  text: {{ $json.brikkorestored.restoredtext }}

This is the most flexible shape — you control which model, which provider,
and you can drop the Restore step entirely if the LLM’s response is just a
classification label that doesn’t need to expose the original PII.

—

Output schema / Схема результата

Each node writes its result under a configurable field (default names below).

Brikko Anonymize → $json.brikko

{
  "maskedtext": "Hi 001>, your order  ships tomorrow.",
  "mapping_id":  "0e1f7a3a-d6e5-4d12-b1e1-a36b66e1c2f0",
  "entities": [
    { "placeholder": "",  "category": "NAME",  "confidence": 0.95 },
    { "placeholder": "", "category": "PHONE", "confidence": 0.99 }
  ],
  "policy":     "balanced",
  "source":     "studio",
  "latency_ms": 14
}

{
  "restored_text": "Hi Ivan Petrov, your order +7 495 123-45-67 ships tomorrow.",
  "hallucinated":  [],
  "source":        "studio",
  "latency_ms":    7
}

{
  "response":         "Здравствуйте, Иван! Спасибо за заявку...",
  "maskedprompt":    "Напиши ответ для 001>...",
  "maskedresponse":  "Здравствуйте, 001>! Спасибо...",
  "mapping_id":       "0e1f7a3a-d6e5-4d12-b1e1-a36b66e1c2f0",
  "model":            "claude-sonnet-4-6",
  "latency_ms":       1840,
  "source":           "gateway"
}

{
  "found_pii": [
    { "category": "EMAIL", "count": 2, "samples": ["a@x.ru", "b@y.com"] },
    { "category": "PHONE", "count": 1, "samples": ["+7 495 123-45-67"] }
  ],
  "total_count": 3,
  "source":      "regex"
}

pnpm install
pnpm run build
pnpm run test
pnpm run lint