Zaprex inspects every prompt, response, and tool invocation across your models in real time — blocking policy violations, redacting PII, and producing an immutable audit trail your legal team will actually trust.
The pipeline
Drop-in SDK or proxy sits in front of OpenAI, Anthropic, Bedrock, or your own model gateway.
Every prompt, tool call, and response is scored against your policies in under 8ms.
Block, redact, or rewrite in real time. Or just observe — you stay in control.
Immutable, cryptographically signed logs your auditors can verify themselves.
Platform
Write policies in plain language or YAML. Hot-reload across every environment without a deploy.
60+ detectors for PHI, PCI, API keys, and custom patterns. Reversible tokens when you need them back.
Detect jailbreaks, indirect injections, and tool-use exfiltration with a model trained on 4M attack samples.
Every call signed and chained. Export to S3, Snowflake, or your SIEM in the format auditors expect.
Edge-deployed inspectors run beside your model, not in a round-trip across the country.
OpenAI, Anthropic, Google, Bedrock, Mistral, Llama, and your fine-tunes. One API, one dashboard.
Compliance
Zaprex ships with mapped controls and evidence collection for the frameworks that matter most. One source of truth, every audit.
Install Zaprex in an afternoon. Have a defensible audit trail by tomorrow morning.
$ npm i @zaprex/guard · 4 lines of code · no model changes