GuardRails

Lightweight content safety scanning for Go LLM applications.

guard := guardrails.New(
    guardrails.WithScanner(scanners.NewPII(), guardrails.ActionRedact),
    guardrails.WithScanner(scanners.NewToxicity(), guardrails.ActionBlock),
    guardrails.WithScanner(scanners.NewPromptInjection(), guardrails.ActionBlock),
)

result := guard.Scan(ctx, userInput)
if !result.Passed {
    // Input blocked
}

Install

go get github.com/promptrails/guardrails

Scanners

Scanner	Detects	Redaction
PII	Email, phone, SSN, credit card, IP address	Yes
Toxicity	Offensive language (keyword matching)	No
BanSubstrings	Custom banned words/phrases	Yes
PromptInjection	Override instructions, jailbreaks, role hijacking	No
Secrets	API keys, tokens, private keys, connection strings	Yes
InvisibleText	Hidden Unicode characters (prompt injection)	Yes
NoRefusal	LLM refusal phrases ("I'm sorry", "As an AI")	No
TokenLimit	Max word count enforcement	No
ReadingTime	Max reading time in seconds	No
JSONValidator	JSON structure + required keys	No
URLReachability	Hallucinated/unreachable URLs	No
BanCode	Code snippets (10 language patterns)	No
MaliciousURL	Suspicious TLDs, IP hosts, phishing keywords	No
Sentiment	Negative sentiment (lexicon-based)	No

Actions

Action	Behavior
ActionBlock	Stop processing, mark as failed
ActionRedact	Replace matches with labels, continue
ActionLog	Record violation, continue unchanged

LLM Guard Integration

For ML-powered scanning, the llmguard sub-package connects to LLM Guard:

import "github.com/promptrails/guardrails/llmguard"

client := llmguard.NewClient("http://localhost:8000", "")

guard := guardrails.New(
    guardrails.WithScanner(
        llmguard.NewScanner(client, "Toxicity", guardrails.ScannerToxicity),
        guardrails.ActionBlock,
    ),
)

Documentation

Getting Started	Installation and quick start
Scanners	All scanners with config options
LLM Guard	ML-powered scanning via LLM Guard API
Custom Scanners	Build your own scanner

Full docs: promptrails.github.io/guardrails

License

MIT — PromptRails

Part of the PromptRails AI Toolkit

• LangRails — Unified LLM provider
• GuardRails — Content safety scanning
• MemoryRails — Agent memory
• MediaRails — AI media generation
• Go AI Toolkit — Demo app