Enterprise AI Firewall & HITL Gateway

Deploy Autonomous AI Agents.
Without the Compliance Nightmare.

Cherubim sits inline between your AI agents and upstream LLMs. Automatically redact PII, block prompt injection, and hold high-risk tool operations for human sign-off.

Start 30-Day Pilot View Integration Code
Proxy Pipeline Architecture
1. AI Agent
Generates Tool Call Payload
2. Cherubim Proxy 👼
Sanitizes PII | Inspects Security | Intercepts High Risk
3. Upstream LLM
Executes Cleared Command

Core Protection Layers

🛡️

Automated PII Redaction

Strips SSNs, credit cards, emails, and sensitive API keys out of prompt payloads before they ever reach cloud-hosted models.

⚔️

Prompt Injection Guard

Scans incoming agent inputs in real-time to intercept adversarial overrides, system prompts leaks, and malicious command injections.

👤

Human-in-the-Loop (HITL)

Intercepts high-risk actions (`execute_payment`, `database_query`) and pauses execution until authorized via live console.

Single Line Integration

Change your base URL. No application refactoring necessary.

agent_config.py ● Docker Proxy Active
# 1. Update target base_url to point to Cherubim client = OpenAI( base_url="http://localhost:5002/v1", api_key=os.getenv("CHERUBIM_AUTH_TOKEN") ) # 2. Agent calls operate normally through security layer response = client.chat.completions.create( model="gpt-4o-mini", messages=[{"role": "user", "content": "Transfer $500 to vendor"}], extra_body={"intended_tool": "execute_payment"} )