Guardrails

elsai Guardrails provides programmable guardrails for LLM-based applications — detect toxic content, protect sensitive data, enforce token budgets, authorize tool calls, defend agent memory, and persist runs through ARMS.

Why elsai Guardrails?

elsai Guardrails provides a comprehensive solution for securing your LLM-based applications. With built-in protection against common threats and flexible configuration options, you can ensure your AI applications are safe and compliant.

Key Features

Comprehensive Protection

Detect and block toxic content, sensitive data, jailbreak attempts, and malicious requests.

Off-Topic Detection

Keep conversations focused by defining allowed topics and blocking off-topic inputs.

SQL Syntax Validation

Validate SQL queries for 7 major dialects before execution to catch errors early.

PII/PHI Detection

Identify sensitive personal and health information with configurable policies, masking, and audit logging.

Token Budget Enforcement

Compute full-context token usage and reject oversized requests before LLM processing.

Tool Authorization

Restrict agent tool access with role-based allowlists, denylists, and sensitive tool gating.

Rate Limiting

Prevent abuse with per-session request, tool call, and execution time limits.

Data Exfiltration Detection

Block or mask LLM responses that leak credentials, bulk PII, or export-style payloads.

Memory Protection

Defend agent memory banks against poisoning with trust scoring, isolation, TTL, and provenance logging.

Agent Trust

Defend multi-agent pipelines with identity/trust registries, HMAC attestation, trust-based privilege isolation, and independent re-validation of upstream agent output.

ARMS Storage

Persist guardrail runs to MongoDB, DynamoDB, or ClickHouse via the ARMS Backend.

Flexible Configuration

YAML-based or programmatic configuration with customizable thresholds and rules.

Easy Integration

Simple Python API with async support, detailed results, and input/output validation.

Multi-LLM Support

Works with OpenAI, Azure OpenAI, Anthropic, Gemini, AWS Bedrock, and more.

Quick Example

from elsai_guardrails.guardrails import LLMRails

# Initialize with configuration
rails = LLMRails.from_config("config.yml")

# Safe LLM calls with automatic guardrails
response = rails.generate(
    messages=[{"role": "user", "content": "Hello!"}]
)

What's New

Version 0.1.8 is a maintenance release with bug fixes. Recent highlights:

• Agent Trust (v0.1.7) — Multi-agent trust boundary with identity/trust registry, HMAC attestation, trust-based privilege isolation, and independent re-validation of upstream agent output

See What's New · Release Notes

Get Started

Ready to secure your LLM application? Check out the Installation Guide or jump to the Quick Start.