Skip to content
AgentsIndex

LLM Guard

Open-source filters for safer prompts and model outputs.

Best for
AI teams who need lightweight security checks around prompts and outputs.
GitHub stars
3.2k

What we checked

Pages last read
Facts checked
1 fact

From 2 sources

AI answers
Too few read

58 buyer-question answers read, November 2026

At a glance

Tags
Open source

Code

Last push
8 July 2026

For developers

Works with
Azure OpenAI, Bedrock, Langchain, GPT, Llama, Mistral and 1 more

What is LLM Guard?

LLM Guard is an open-source security library for LLM applications: it runs a pipeline of scanners over prompts before they reach a model and over outputs before they reach a user, covering prompt injection, PII anonymization and deanonymization, secrets, toxicity and banned topics.

It runs on CPU, is model-agnostic, and is documented for OpenAI, LiteLLM Proxy, LangChain and LlamaIndex, as a library or a self-hosted API server. The repository was archived in July 2026, so what exists today is a complete but frozen MIT-licensed codebase rather than a project in development.

Archived in July 2026, and where the work went

The protectai/llm-guard repository was archived by its owner on 9 July 2026 and is now read-only; GitHub's API confirms archived: true and the last commit landed 8 July 2026, the day before (checked 16 October 2026). The project's own README now carries a line added at the same time: "This project and its associated models on Hugging Face are no longer under active development or maintained." This isn't an abandoned side project — it's a wind-down with a clear cause. Palo Alto Networks completed its acquisition of Protect AI, the company behind LLM Guard, on 22 July 2025; Protect AI's CEO joined Palo Alto Networks as VP of Product for Prisma AIRS, and the company's own announcement describes Protect AI's technology as becoming "a cornerstone of" that commercial platform. LLM Guard kept shipping for roughly a year under its new owner — its last PyPI release, 0.3.16, went out 19 May 2025, before the deal closed — and was then frozen as the security work consolidated into Prisma AIRS, which is a separate, commercial product we have not evaluated here. No fork has picked up where it left off: the five most recently updated forks on GitHub all show the same last-commit timestamp as upstream, meaning none has diverged.

What the code actually does

LLM Guard is a Python library that runs a pipeline of scanners on both sides of an LLM call: on prompts before they reach a model, and on outputs before they reach a user. Its input scanners cover prompt injection and jailbreak attempts, PII anonymization (with a matching deanonymize step to restore masked values once the model has replied), secrets, banned topics and competitors, code detection, gibberish, and sentiment; output scanners add toxicity and bias scoring, malicious or unreachable URL detection, and relevance/factual-consistency checks. It ships as either an importable library or a self-hosted API — the docs describe deploying it with FastAPI and Uvicorn — and its classifier models are small enough to run on CPU. It's model-agnostic: documented integrations cover calling OpenAI directly, sitting behind LiteLLM Proxy to moderate calls across providers like Anthropic, Bedrock and Gemini, and wiring into LangChain or LlamaIndex retrieval pipelines.

Residual usage, checked today

Despite the freeze, the package still moves: PyPI shows 106,172 downloads of the llm-guard package in the 30 days to 16 October 2026, and its prompt-injection classifier on Hugging Face (protectai/deberta-v3-base-prompt-injection-v2) logged 690,976 downloads over the same window. The GitHub repo itself carries open issues that will now never be closed by a maintainer, since archived repos on GitHub don't accept new issues or pull requests. That combination — real, ongoing pickup against a project nobody is patching — is the actual trade-off here, not a reason to dismiss it outright.

What you inherit by adopting it now

The code is MIT-licensed, so there's nothing stopping a team from forking it, patching it, and running it indefinitely — but that work is now entirely on the adopter. There will be no future fixes for scanner bugs, no refreshed detection models as new prompt-injection or jailbreak techniques appear, no new framework integrations, and no vendor security advisories beyond what's already in the repo's SECURITY.md. The project's own domain, llm-guard.com, which its README still links to for documentation and changelog, no longer resolves (checked 16 October 2026); the GitHub Pages mirror at protectai.github.io/llm-guard/ is the only live copy of the docs today.

What we checked on LLM Guard

Questions about LLM Guard

Is LLM Guard open source, and can you host it yourself?
LLM Guard is open source. We checked github.com/protectai/llm-guard/blob/main/LICENSE ourselves on 15 November 2026.
What have we checked on LLM Guard?
1 fact, under one heading: How you build with it.
How do we check LLM Guard?
We hold 1 fact on LLM Guard, quoted from 2 pages. Facts last checked 15 November 2026. Last change found 15 November 2026.

How we check every listing

Sources and changes

Last full check

What changed1

  • What it does

Repository

Licence
MIT
Stars
3,216
Last push
8 July 2026

Similar in Agent tools and integrations

See all in Agent tools and integrations

From our research

All research

Is LLM Guard yours?

Free, and the page then carries the verified owner mark. Claim this listing

Report an error if a fact is wrong.