EVRGUARD API

Check what your AI agent is about to do, before it does it.

EvrGuard screens each proposed agent action against your rules and returns a verdict.

Wrap a tool with the SDK and a refused action never executes.

Get Started FreeView the docs

96.6% hostile catch rate

56 of 58 hostile actions caught, with zero false positives, on an internal 73-case adversarial suite. Measured April 2026.

Works with any AI agent.
Claude CodeClaude DesktopOpenClawLangChainLangGraphAutoGenCrewAIAny Python Agent
How It Works

Three verdicts

STEP 1

CONNECT

PASS — no rule matched. The action runs.

STEP 2

VERIFY

STEERED — a rule matched. The call is stopped and the agent is given specific guidance on a compliant alternative, citing the rule it breached.

STEP 3

ENFORCE

BLOCKED — the action is refused outright. No alternative is offered.

main.py
1
2
3
4
5
6
7
8
from evrguard import EvrmindGuard
guard = EvrmindGuard(api_url="https://api.evrmind.io")
@guard.wrap_tool
def execute_query(sql):
return db.execute(sql)

Every proposed action is checked against your configured rules. Screening returns a clear verdict with the matched rule and reason. Paid plans add intent analysis when the pattern layers find nothing.

Intelligent Safety

Where guidance helps

When an action violates a rule, EvrGuard stops the call and gives the agent specific guidance on a compliant alternative, citing the rule it breached.

1
Agent Proposes
"Export customer records with home addresses to external partner"
EvrGuard Verdict: STEERED
2
Safe Alternative
"Export anonymised customer analytics to internal reporting system"
Agent continues working
Integration Methods

Three ways to connect

Python SDK

Wrap any function. Works with LangChain, LangGraph, AutoGen, CrewAI, or custom agents.

pip install evrguardView on GitHub

MCP Server

Zero code integration for Claude Code, Claude Desktop, and OpenClaw. Add one config line. Every tool call is verified automatically.

pip install evrguardSetup guide

HTTPS Proxy

Works with any language, any application. Set your proxy variable. All HTTP traffic passes through EvrGuard.

pip install evrguardSetup guide
Features

Everything you need for agentic safety

Triple-Layer Detection

Every action passes through all three detection layers automatically. Pattern matching runs first. The engine safety net catches what patterns miss. Intent analysis catches the rest. One product. Three layers. No configuration needed.

Rule-based guidance

When a rule matches, the call is stopped and the agent receives specific guidance on a compliant alternative, citing the rule it breached.

Auditable verdicts

Every decision records the rules checked, the verdict returned, and the reason. Exportable logs support review and incident analysis.

Industry rule library

Start with practical rule packs for finance, healthcare, legal, DevOps, retail, education, insurance, government, and critical infrastructure.

Regulatory rule packs

Rule packs cite the article or section they come from, so a blocked action tells you which obligation it touched.

Monitor Mode

Run alongside your agents without blocking anything. See what would be caught. Build confidence in your rules. Switch to enforce when ready.

Rule packs

Pre-built rules for common workflows

Select a rule pack to preview the checks your workflow can apply. Packs are configurable and cite the source rule in each verdict.

compliance.evr
Select a framework to see its enforcement rules

Plus: DORA, NIS2, CCPA, COPPA, FERPA, PCI-DSS, and industry-specific constitutions for 10 sectors.

*Compliance packs provide technical enforcement infrastructure and do not constitute legal advice. Review rules with legal counsel for your specific use case.

Constitutional Compiler

Write your own rules. Or choose from our library. Enforce everywhere.

Define your safety rules in a simple, human-readable constitution. Our compiler translates them to both software verification and hardware enforcement.

my-constitution.evr

Software

EvrGuard API

Mathematical verification in <1ms

Audit evidence

A clear record of the rule, verdict, and reason for every screened action.

Pricing

Simple, transparent pricing

Free

£0/month
  • 1,000 verifications per month
  • Choose from 10 templates
  • Documentation and guides
Get Free API Key

Starter

£69/month
  • 10,000 screenings
  • Custom rule packs
  • Webhook alerts
  • Email support
Subscribe
Most Popular

Professional

£279/month
  • 100,000 screenings
  • Rule packs
  • Exportable audit logs
  • Usage dashboard
  • Priority support
Subscribe

Enterprise

Custom
  • Unlimited screenings
  • Dedicated instance
  • SLA
  • Custom rule packs
  • Auditor portal
The Core Difference

How enforcement works

Prompt Guardrails

Asks the AI to behave. Defeated by jailbreaks.

AI Safety Layers

Uses AI to check AI. Same failure modes.

EvrGuard

Rules, evidence, and an explicit decision for every proposed action.

97.3% hostile catch rate. Every screened action leaves an explainable verdict your team can review.

Ready to secure your AI agents?

Start with 1,000 free verifications per month. No credit card required.

Get Started Free