On-Premises · BYOK · §203-compliant

On-premises AI for regulated professions.

Saklam Bridge masks PII inside your network and forwards only tokens to Anthropic, OpenAI, Google, Azure, AWS, Mistral, or your own inference cluster. Client data never leaves your premises in plaintext.

Book a 30-min demo

How it works

What Saklam sees: nothing. What the LLM provider sees: masked tokens.

Workstation

Claude Code, Cursor, or your own apps point at bridge.intern.firm.com

Saklam Bridge (Docker)

Runs on-prem inside your firm network. Masks PII (453 patterns) before requests leave the house.

LLM Provider

Anthropic, OpenAI, Google, Azure, AWS Bedrock, Mistral, or your self-hosted endpoint see only tokens like [PER_a1b2]. The response is unmasked on the way back.

Three properties, three sales arguments

Zero-Knowledge

Saklam servers are not in the data path. The container runs on-prem, sees all plaintext — Saklam sees nothing. Real technical separation, not just contractual assurance.

§203-compliant

Client data never leaves your premises as plaintext. No DPA with Saklam needed — you sign that directly with Anthropic, OpenAI, or Mistral.

BYOK

Anthropic, OpenAI, Google, Azure OpenAI, AWS Bedrock, Mistral, or your own inference cluster (Ollama/vLLM) — your direct provider relationship. We don't mark up your AI usage; we license you the software.

Live in 5 minutes

Linux x86_64 or ARM64, 2 vCPU + 4 GB RAM minimum. Docker Engine ≥ 24.

1. Create directory 
mkdir -p /opt/saklam-bridge && cd /opt/saklam-bridge
2. Download docker-compose.yml + .env template 
curl -fsSL https://raw.githubusercontent.com/saklam/bridge/main/docker/docker-compose.yml -o docker-compose.yml
curl -fsSL https://raw.githubusercontent.com/saklam/bridge/main/docker/.env.example -o .env
3. Enter ANTHROPIC_API_KEY and BRIDGE_MASTER_KEY into .env 
$EDITOR .env
4. Start 
docker compose pull && docker compose up -d
5. Smoke test 
curl -fsS http://localhost:4000/health/readiness

Full setup guide →

Pricing

Licensed per Named User (= staff regularly using Bridge). Quarterly cutoff.

Pilot

30 days free
1-3 lighthouse customers
  • Unlimited volume
  • Setup support directly with Stefan
  • Feedback flows into the roadmap
Request pilot

Pro

€299
/ user / year
  • Auto-updates
  • Email support within 2 business days
  • DPA addendum as PDF
Book demo

Enterprise

€499
/ user / year
  • Priority support (4h within business hours)
  • Custom PII patterns
  • Audit trail export
Book demo

Cost example: 60-staff firm on Pro tier = €17,940/year. Compare: 60 × Libra Pro at €200/mo = €144,000/year.

Frequently asked questions

Do we need a DPA (Auftragsverarbeitungsvertrag) with Saklam?

No. Saklam Bridge runs on-prem inside your infrastructure, Saklam receives no personal data. A DPA under Art. 28 GDPR would be content-empty. Instead we provide a site-license agreement + data protection addendum for your DPO documentation. A DPA with the LLM provider (Anthropic, OpenAI, Mistral) is still required and you sign that directly with the provider — we don't resell it through us.

Which LLM providers are supported?

Anthropic (Claude), OpenAI (GPT), Azure OpenAI (same GPT models with EU data residency + Microsoft DPA), Google Gemini (Vertex AI EU or AI Studio), AWS Bedrock (Claude/Llama/Mistral via AWS, EU Frankfurt), Mistral La Plateforme, and self-hosted via Ollama/vLLM/TGI inside your own network. Customer-provider relationship is direct — Saklam is not a reseller.

We run our own inference cluster (Ollama/vLLM). Does Bridge add value there?

Yes. Point Bridge at your cluster via OLLAMA_API_BASE (OpenAI-compatible API). The PII masking layer stays active — even with on-prem inference you get consistent token formats, an audit trail, and unified routing logic for both your local models and cloud fallbacks. Belt-and-suspenders.

We use Claude Pro / Anthropic Team / ChatGPT Plus. Does that work?

Unfortunately not. Bridge only works with classic API keys (pay-as-you-go, Console → API Keys). Subscription OAuth tokens are not accepted.

What hardware do we need?

Linux x86_64 or ARM64. 2 vCPU + 4 GB RAM minimum, 4 vCPU + 8 GB recommended for multi-user. ~3 GB storage for container + models.

What is the latency overhead?

~70 ms PII detection overhead per request (Linux CPU). For structured PII (Email/IBAN/phone) <100 ms; for PERSON/ORG detection (GLiNER + Presidio) 1-2 s additional on first request, then cached.

How does unmasking work with streaming responses?

Bridge unmasks chunk-by-chunk, including tokens that split across chunks. Works for both OpenAI and Anthropic SSE streams.

Can we add custom PII patterns for our client workflows?

Enterprise tier includes this. Pro tier on request.

How do we update the image?

docker compose pull && docker compose up -d. Auto-updates are off — you control what runs in your network.

Is Saklam Bridge open source?

Closed-source, proprietary. Image on hub.docker.com/r/saklam/bridge, setup docs public, license: site license per Named User.

Looking for pilot customers

We're looking for 1-3 mid-sized IT-aware law firms (30-100 staff) for a 30-day pilot. Unlimited use, personal setup support from the founder, feedback flows directly into the roadmap. Afterwards: standard site license.

Book a 30-min demo

Direct questions? bridge@saklam.com