tamga-proxy intercepts every app→model request, runs PII, secret, and injection scanners inline under 5ms p95, and enforces policy actions — before a single token reaches the provider.
One Docker Compose command. Zero architecture changes. Your existing LLM SDK calls route through Tamga transparently.
Change your base_url. That's the only code modification required. No SDK wrappers, no middleware.
Unprotected prompt vs. Tamga-protected request. PII redacted, injection blocked, audit logged.
Potential leak: raw PII reaches provider.
Paste a prompt, run the policy engine, and inspect the findings in real time.
Write the policy in YAML and watch the action and redaction result land on a sample prompt live.
| type | category | rule | action |
|---|---|---|---|
| pii | credit_card | redact-pii-tr | REDACT |
| pii | redact-pii-tr | REDACT | |
| pii | tc_kimlik | redact-pii-tr | REDACT |
| pii | phone_tr | redact-pii-tr | REDACT |
Threat-hunting sandbox: payload inspector, analysis meter, and findings table.
Findings
| Severity | Finding Type | Data (Redacted/Key) | Risk Score | Action Taken | Confidence |
|---|---|---|---|---|---|
| CRİTİCAL | pii:credit_card | 4532…0366 | 96% | BLOCK | 0.95 |
| HİGH | pii:email | ayse….com | 84% | REDACT | 0.89 |
| CRİTİCAL | pii:tc_kimlik | 1000…0146 | 96% | BLOCK | 0.95 |
| HİGH | pii:phone_tr | +90 …4567 | 84% | REDACT | 0.89 |
{
"redacted_payload": "Customer: [REDACTED_EMAIL]\nTC: [REDACTED_TCKN]\nCall me at [REDACTED_PHONE]\nCard: [REDACTED_CC]"
}Verified Performance
Every benchmark is reproducible from a single go run ./cmd/redteam command against a public 309-prompt adversarial corpus. No cherry-picking, no marketing dataset.
Tamga detects and mitigates every category in the OWASP LLM Application Security framework.
Tamga streams every security event to your SIEM, alerting, and logging infrastructure. No rip-and-replace — just add a webhook.
Tamga is designed for regulated industries. Banking, healthcare, and government teams can deploy with confidence.
Local processing with zero data retention. No PII leaves your infrastructure. All redaction happens inline before any external API call.
Real-time credit card number detection via strict Modulus 10 (Luhn) validation. PAN data is redacted before it reaches the LLM provider.
Cryptographically verifiable event logging to PostgreSQL. Every prompt, redaction, and policy decision is timestamped and queryable via SQL.
Complete control over outbound LLM API traffic. No model sees your data unless your policy explicitly allows it. Block shadow AI providers by default.
PRICING //
From self-hosted open-source to air-gapped enterprise — one pricing model that scales with your AI security maturity.
For individuals and small teams getting started with LLM security.
Managed Tamga cloud for development teams up to 50 engineers.
For production AI workloads with compliance and SLA requirements.
Air-gapped deployment, custom compliance, dedicated engineering.
All prices in USD. Volume discounts available for 50+ developers. Contact sales for custom pricing.
GET STARTED //
Deploy Tamga in front of your LLM APIs. No SDK changes. No application code rewrite. Full visibility in under 5 minutes.
No credit card required · Self-hosted option available · 5-minute setup
Need enterprise pricing? Talk to sales →