Launching soon — PromptGuard is open source on GitHub

Zero Trust for AI

Every prompt inspected. Every agent action monitored. Every risky command caught before it executes.

Register Interest View on GitHub
Employee AI Agent Application RAUCLE Pattern 180+ rules <1ms Semantic ML / Heuristic <10ms OpenAI Anthropic Azure Bedrock YOUR ORG DETECTION PIPELINE AI PROVIDERS Prompt injection blocked — PI-001, PI-004 Agent scope violation — rm -rf blocked — MCP-002 Clean prompt — forwarded to model
180+
Open Source Detection Rules
<1ms
Pattern Matching
6
AI Providers Supported
0
Mandatory Dependencies
Platform

Complete AI security
in a single gateway

Everything you need to secure AI usage across your organisation.

AI Gateway

Reverse proxy that sits between your org and every AI provider. One endpoint for OpenAI, Anthropic, Azure, and Bedrock. Streaming SSE with under 10ms overhead.

Prompt Injection Detection

Three-layer inspection pipeline. 180+ open-source detection rules ship free with PromptGuard, plus a proprietary rule set developed by Raucle available exclusively in the commercial platform.

Data Loss Prevention

Detects PII, credentials, API keys, NI numbers, NHS numbers, and IBANs. Redacts or blocks before sensitive data ever reaches an AI model.

Shadow AI Discovery

Discover which AI tools are in use across your organisation and which departments use them. Sanction, review, or block any tool.

AI Agent Security

Monitor MCP agent tool calls, data access, and scope violations. Detect tool poisoning attacks and block compromised agents in real time.

Policy Engine

YAML-based rules with per-user, per-department, per-tool granularity. Test rules against historical data before activation. Full audit trail.

Architecture

How it works

No prompt reaches an AI model and no agent action executes without being inspected first.

INSPECTION PIPELINE User / Agent RAUCLE Gateway Pattern 180+ rules <1ms Policy Allow / Alert / Block AI Model BLOCKED Prompt injection PI-001 · instruction_override Confidence: 0.89 · Action: BLOCK SOC ALERTED Agent scope violation deploy-bot · rm -rf /var/data Escalated to SOC · Ticket #4821 CLEAN Forwarded to model Logged to SIEM SIEM INTEGRATION Splunk Elasticsearch Sentinel SOC Console BLOCKED ALERT ALLOW
Agent Security

See every action
your AI agents take

AI agents don't just process prompts — they execute tool calls, access files, run commands, and make API requests. Raucle monitors every action in real time and blocks risky behaviour before damage is done.

  • Dangerous commands blocked Shell execution, file deletion, and privilege escalation caught before they run
  • Data exfiltration stopped Agents attempting to send credentials, PII, or secrets to external endpoints
  • Scope violations detected Tool calls outside the agent's approved permissions flagged and blocked
  • Tool poisoning caught MCP tool descriptions altered after approval detected and quarantined
Agent Activity Monitor
Live — 3 agents active
Shell command blocked Agent deploy-bot attempted rm -rf /var/data/* — destructive operation outside scope
Blocked
Credential exfiltration stopped Agent research-agent tried to POST AWS_SECRET_KEY to external webhook
Blocked
Tool description changed MCP tool file_reader manifest modified after approval — possible rug pull
Alert
Privilege escalation attempt Agent code-assistant called sudo chmod 777 — elevated permissions denied
Blocked
Unusual data access pattern Agent analytics-bot reading 847 customer records in 3 seconds — rate limit triggered
Alert
Tool call approved Agent support-agent called search_knowledge_base — within scope, clean prompt
Allowed
Prompt injection via tool input Agent email-drafter received injected instructions in customer email body — PI-001, PI-005
Blocked
Open Source

PromptGuard
detection engine

The core detection engine that powers Raucle is available as a standalone MIT-licensed Python library with 180+ open-source detection rules. The commercial platform extends this with a proprietary rule set developed by Raucle, covering advanced evasion techniques, zero-day attack patterns, and industry-specific threats.

  • 180+ open-source detection rules included
  • Zero mandatory dependencies — pip install and go
  • Python library, CLI, and REST API
  • Extensible YAML rules — bring your own or use ours
  • Optional ML classifier with transformers

The Raucle commercial platform includes additional proprietary rules not available in the open-source edition.

Star on GitHub
terminal
$ pip install promptguard $ python from promptguard import Scanner scanner = Scanner() result = scanner.scan("Ignore all previous instructions") print(result.verdict)    # "MALICIOUS" print(result.confidence) # 0.8925 print(result.action)     # "BLOCK" # Clean prompts pass through result = scanner.scan("What is the capital of France?") print(result.verdict)    # "CLEAN"
Compliance

Built for UK & EU compliance

Full enforcement of the EU AI Act begins August 2026. Raucle is ready.

EU AI Act Ready

Compliant with high-risk AI system requirements

GDPR Compliant

PII detection, data processing agreements, and right-to-erasure

UK Data Residency

Data stays in the UK. No transatlantic transfers required

SOC 2 Conformant

Audit logging, access controls, and continuous monitoring aligned to SOC 2 principles

Stop AI threats
before they start

Be first to know when Raucle launches. Get early access to the platform that stops prompt injection, data leaks, and rogue AI agents.

Register Interest Explore PromptGuard