LLM Guard alternative after the archive: Nonimo Gateway

LLM Guard did the whole PII round trip for LLM calls: it anonymised the prompt, kept the real values in a vault and put them back into the answer. Its repository is now archived, and the project says it is no longer maintained. If you depend on it, you need a replacement that is actively developed.

Nonimo Gateway runs on your own servers. See plans and get in touch.

This guide sets out what the archive means in practice, where LLM Guard falls short for European documents, and how Nonimo Gateway takes over the PII part of the job.

What happened to LLM Guard

LLM Guard, “The Security Toolkit for LLM Interactions”, was built by Protect AI as a set of input and output scanners. It offered a documented, reversible loop: Anonymize replaced personal data with placeholders, a Vault kept the originals, and Deanonymize “helps put back real values in the model’s output by replacing placeholders”, alongside scanners for prompt injection, secrets and toxicity.

Two things have changed:

• The repository is archived. The README opens with “THIS PROJECT HAS BEEN ARCHIVED.” and says the project and its models on Hugging Face “are no longer under active development or maintained.” The last release on PyPI is 0.3.16, uploaded on 19 May 2025.
• Protect AI is part of Palo Alto Networks. The acquisition was completed on 22 July 2025, with Protect AI’s technology described as “a cornerstone of” Prisma AIRS. Palo Alto Networks also closed the acquisition of Portkey on 29 May 2026, so the two now share an owner. See our Portkey alternative guide.

Where LLM Guard falls short today

Nobody upstream maintains it. Your container still starts, but nobody will patch the code, its dependencies or its models. If a client or an insurer asks how you keep personal data out of AI tools, “an archived open-source project” is an answer you will have to defend. Our guide to the AI questions on cyber insurance forms shows the wording you may be asked to sign.

English-specific detection. The Anonymize page says “The scanner can detect PII in English and Chinese,” and that “Current entity detection functionality is English-specific.” Its default list has ten types, such as names, emails, credit cards, US Social Security numbers and IBANs. A Spanish NIE, an Italian codice fiscale or a German tax ID is not among them. The scanner can also draw on Presidio’s recognisers and your own regex patterns, but the tuning for other languages is then yours, on a project nobody maintains.

A heavy container. The API deployment page recommends “at least 16GB of RAM allocated to Docker.”

No proxy. The pages we checked do not document an OpenAI-compatible proxy, streaming or tool calls. LLM Guard is a scanning service your application calls before and after its own call to the model, so the streaming restoration is yours to build.

LLM Guard compared with Nonimo Gateway

	LLM Guard	Nonimo Gateway
Status	Archived, no longer maintained	Available
How you call it	Scanning library or API, before and after your own LLM call	OpenAI-compatible proxy, you change base_url
Restores real values	Yes, Vault and Deanonymize	Yes, real values back in the answer
Streaming	Not documented	Real values back in the answer
Languages and IDs	English-specific; ten default entity types	National documents per country, such as Spanish DNI/NIE/NIF, German Steuer-ID, French NIR, Italian codice fiscale, Brazilian CPF, Mexican RFC and CURP
Resources	At least 16 GB of RAM recommended for Docker	Runs on your own servers
Other scanners	Prompt injection, secrets, toxicity and more	PII only
Licence	MIT, open source	Plans on the Nonimo Gateway page

If you are weighing other routes, Presidio is the maintained detection layer LLM Guard was built on, though not a proxy, and LiteLLM PII guardrails covers masking inside LiteLLM, where Nonimo also has a plugin.

How Nonimo Gateway does it

Nonimo Gateway covers the PII part of what LLM Guard did, as a proxy rather than a library. It runs on your own servers. You point your client at it and keep your code:

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://llm-gateway.your-company.internal/v1",  # your Nonimo Gateway
    api_key=os.environ["OPENAI_API_KEY"],  # your own OpenAI key
)

It replaces personal data in what goes out and restores the real values in the response. It recognises the national documents and numbers of each country it serves, including Spain’s CUPS energy supply number alongside the national IDs above. It uses your own OpenAI key, and there is a plugin for LiteLLM. Plans and prices are on the Nonimo Gateway page. It does not replace LLM Guard’s prompt injection or toxicity scanners.

This is pseudonymisation, and pseudonymised data is still personal data for whoever holds the mapping; our guide on de-identified versus anonymised data explains the difference. For the written rule your staff follow, our AI policy template is a starting point.

Replace it before it ages further

If Anonymize and Deanonymize are the scanners you actually run, you do not need to carry an archived toolkit. Nonimo Gateway does that round trip on your own server. See Nonimo Gateway’s plans and get in touch.

Sources

Checked 20 October 2026.

• LLM Guard, GitHub repository. Archive notice, scanner list, MIT licence.
• LLM Guard, Anonymize scanner. Entity list, languages, Vault.
• LLM Guard, Deanonymize scanner. Restoring real values in the output.
• LLM Guard, API deployment. Docker image, 16 GB RAM recommendation.
• LLM Guard, API overview. Deployment as an API.
• Palo Alto Networks, Protect AI acquisition completed. Date and Prisma AIRS role.
• Palo Alto Networks, Portkey acquisition completed. Date of closing.

---

This page is for information. Product details of other vendors were checked on their official pages on 20 October 2026 and can change.