# Llama Guard

- **Canonical URI:** https://www.vioscale.ai/software/llama-guard
- **Category:** AI Guardrails
- **Homepage:** https://github.com/meta-llama/PurpleLlama
- **Also known as:** llama-guard
- **Profile claimed by vendor:** no
- **Last updated:** 2026-08-26T19:33:55.962Z

## Vioscale score

**53.3 / 100**, confidence 27% (low). Computed 2026-09-01.

Composite of weighted, independently-sourced signals (no user reviews, no vendor payment).

| Signal | Score | Weight | Contribution | Evidence present |
|---|--:|--:|--:|:--:|
| price_level | 100 | 0.052 | 5.2 | ✓ |
| reliability | 0 | 0.073 | 0 | - |
| capabilities | 57.8 | 0.084 | 4.9 | ✓ |
| repo_stars | 68.7 | 0.026 | 1.8 | ✓ |
| integrations | 0 | 0.09355555555555557 | 0 | - |
| dependent_projects | 0 | 0.063 | 0 | ✓ |
| dev_activity | 28.4 | 0.094 | 2.7 | ✓ |
| release_cadence | 0 | 0.052 | 0 | - |
| security_posture | 0 | 0.073 | 0 | - |
| package_downloads | 0 | 0.136 | 0 | - |
| security_score | 59 | 0.042 | 2.5 | ✓ |
| pricing_transparency | 80 | 0.084 | 6.7 | ✓ |
| community_qa_activity | 0 | 0.063 | 0 | - |

## Pricing

_As of 2026-08-21, [verify at source](https://github.com/pricing). Independently observed._

Open source · Free tier

> Free and open source under MIT license

## About

A suite of input/output moderation models and security evaluation benchmarks designed to help developers build safer generative AI applications. Includes tools to detect prompt injection, jailbreaking attempts, and other violating content.

_Independently observed._

## Platform & deployment

- **Platforms:** CLI
- **Deployment:** Self-hosted

## Capabilities

_The capabilities that matter for AI Guardrails. "-" = undocumented, not absent._

| Capability | Supported |
|---|:--:|
| **Capabilities** | |
| Architecture model | Open source python library |
| Prompt injection and jailbreak blocking | ✓ |
| Inbound pii and credit card masking redaction | ✗ |
| Outbound toxicity and competitor mention blocking | ✓ |
| Hallucination and topical off brand drift detection | ✗ |
| Sub 50ms latency guarantees for proxy routing | ✗ |
| Custom regex and deterministic denylist rulesets | ✗ |
| Streaming token interception and chunked eval | ✗ |
| Rag retrieval chunk poisoning defense | ✗ |
| Zero data retention and prompt privacy guarantees | ✓ |
| SOC2 type ii | ✗ |
| ISO 27001 | ✗ |
| Pricing model | Free open source |

## Security & compliance


Known vulnerabilities: 0 (0 in the last 12 months) ([source](https://advisories.ecosyste.ms/api/v1/advisories?ecosystem=go&package_name=github.com%2Fmeta-llama%2FPurpleLlama&per_page=100)). A count reflects scale and disclosure, not quality.

## Facts

Every value below carries its source and our confidence. Facts are re-crawled on a freshness schedule.

### pricing

| Attribute | Value | Source | Retrieved | Confidence |
|---|---|---|---|---|
| pricing.model | commercial | [link](https://github.com/meta-llama/PurpleLlama) | 2026-08-26 | 70% (medium) |
| pricing.free_tier | yes | [link](https://github.com/pricing) | 2026-08-21 | 60% (medium) |
| pricing.price_level | free | [link](https://github.com/pricing) | 2026-08-21 | 60% (medium) |
| pricing.transparent | yes | [link](https://github.com/pricing) | 2026-08-21 | 60% (medium) |

### activity

| Attribute | Value | Source | Retrieved | Confidence |
|---|---|---|---|---|
| activity.commits_last_30d | 7 | [link](https://github.com/meta-llama/PurpleLlama/pulse) | 2026-08-26 | 65% (medium) |

### adoption

| Attribute | Value | Source | Retrieved | Confidence |
|---|---|---|---|---|
| adoption.github_stars | 4,364 | [link](https://github.com/meta-llama/PurpleLlama) | 2026-08-26 | 90% (high) |
| adoption.dependent_repos | 0 | [link](https://packages.ecosyste.ms/api/v1/packages/lookup?repository_url=https%3A%2F%2Fgithub.com%2Fmeta-llama%2FPurpleLlama) | 2026-08-26 | 85% (high) |

### language

| Attribute | Value | Source | Retrieved | Confidence |
|---|---|---|---|---|
| language.primary | Python | [link](https://github.com/meta-llama/PurpleLlama) | 2026-08-26 | 90% (high) |

### security

| Attribute | Value | Source | Retrieved | Confidence |
|---|---|---|---|---|
| security.scorecard | 5.9 | [link](https://api.securityscorecards.dev/projects/github.com/meta-llama/PurpleLlama) | 2026-08-26 | 90% (high) |
| security.vulnerabilities | `{"count":0,"source":"https://advisories.ecosyste.ms/api/v1/advisories?ecosystem=go&package_name=github.com%2Fmeta-llama%2FPurpleLlama&per_page=100","last_12m":0,"max_severity":null}` | [link](https://advisories.ecosyste.ms/api/v1/advisories?ecosystem=go&package_name=github.com%2Fmeta-llama%2FPurpleLlama&per_page=100) | 2026-08-26 | 90% (high) |

---
*Source: Vioscale (https://www.vioscale.ai/software/llama-guard). Independent, evidence-based software intelligence. Cite the canonical URI.*