Docker · BYOK · Zero-knowledge

PII masking for your LLM pipeline — as a container on your turf.

Saklam Bridge runs inside your infrastructure and strips names, emails, IBANs & 300+ vetted PII patterns from prompts before they ever reach OpenAI, Anthropic or Mistral — and un-masks the response on the way back. Drop-in, deterministic, nothing leaves your network in plaintext.

Setup docs How it works

# Mask PII before it ever reaches an LLM — inside your network
curl http://localhost:4000/v1/mask \
  -H "Content-Type: application/json" \
  -d '{"text":"Mandant Hans Müller, IBAN DE89..."}'

# → "Mandant [PER_a1b2c3d4], IBAN [IBA_c3d4e5f6]"
#   mapping returned for deterministic un-masking
                

One container, two interfaces

Use the stateless mask REST API for documents and pipelines — or put the LLM gateway in front of your existing calls with one base URL.

LIVE

Mask REST API

Stateless. POST /v1/mask in, masked text + mapping out — straight from your own container.

300+ vetted patterns, EU-27 + EEA, bidirectional
No storage — no logs, no database
Runs entirely inside your infrastructure

API examples

LIVE

LLM Gateway

Anthropic- & OpenAI-compatible drop-in. Point ANTHROPIC_BASE_URL / OPENAI_BASE_URL at your container — masking happens transparently.

One env var — works with Claude Code, Cursor & your own apps
BYOK — your provider keys, your bill
Mask on the way out, un-mask on the way back — streaming included

See the quickstart

Quickstart

docker compose up, one base URL changed. PII never leaves your network in plaintext.

# 1. Run the bridge in your infra (your provider keys via .env)
docker compose up -d   # saklam/bridge:latest

# 2. Point your client at it — PII is masked on the way out,
#    un-masked on the way back. Same response shape.
client = OpenAI(api_key=KEY, base_url="http://localhost:4000/v1")

# Claude Code / Anthropic SDK: one env var
export ANTHROPIC_BASE_URL=http://localhost:4000
                

Ship AI features without exposing customer data.

300+ vetted EU PII patterns + multilingual NER, zero-knowledge towards Saklam, 30-day free trial — built by a solo founder who uses it daily himself.

Setup docs on GitHub See pricing