Limina alternative: Private AI pricing and a proxy you run

Private AI pricing has one published number: a free Starter tier of up to 75 API calls a day. The Batch and Enterprise tiers are priced on request. You will now find that page at getlimina.ai, because Private AI renamed itself Limina on 5 March 2026.

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

This guide sets out what Limina publishes about price and the round trip to a large language model, and where a proxy in front of the OpenAI API, such as Nonimo Gateway, fits instead. Every vendor detail links to the vendor’s own page.

Private AI is now Limina

The company announced the change in a post dated 5 March 2026. The post states that the rebrand does not affect its products, customers or operations. In practice, www.private-ai.com redirects to www.getlimina.ai/en, and the documentation is served as Limina Docs. If you searched for “private ai” or “limina”, it is the same product.

Private AI pricing: what Limina publishes

The pricing page states that all prices are in US dollars and billed monthly.

Tier	Published price	What the page says
Starter	Free	Up to 75 API calls a day, 10 calls a minute; file limits of 250 KB for txt, json, csv and xml, 4 MB for other types
Batch	Volume pricing	Quoted on request
Enterprise	Custom pricing	Committed volume, no consumption billing

So a developer can test the API at no cost, and a production budget starts with a sales conversation. Treat Starter as an evaluation tier: an integration that routes every prompt from an office through it can use up 75 calls well before the end of a working day.

Limina deploys as a container in your on-premises environment or VPC, is sold through the AWS, Snowflake and Azure marketplaces, and has a SaaS version where data is processed in Limina’s cloud. Its published hardware minimum for the CPU container is 7.5 GB of free RAM and a 50 GB disk; the GPU container needs 28 GB of RAM and an NVIDIA GPU with at least 16 GB of VRAM.

How Limina puts the real values back

Limina covers text, files, images and audio in 52 languages, according to its homepage.

For a chat round trip, it documents a re-identify route, POST /process/text/reidentify, and a guide to LLM prompt redaction. The pattern is three calls, made by your code in order:

1. your app  -> Limina   de-identify the prompt
2. your app  -> OpenAI   send the de-identified prompt
3. your app  -> Limina   re-identify the answer

That is where it may not fit a small team:

• It is an API you call, not a proxy. The documentation index has no page for an OpenAI-compatible endpoint, so every place in your code that calls a model needs two extra calls wrapped around it.
• Streaming is not documented. If your application streams answers token by token, the docs do not say how to re-identify as of 20 October 2026.
• Paid prices are not published. Above 75 calls a day, you need a quote.

A reversible process is pseudonymisation, not anonymisation. Our guide on deidentified versus anonymised explains why that matters to a regulator.

Limina and Nonimo Gateway side by side

	Limina (formerly Private AI)	Nonimo Gateway
What it is	Container-based de-identification service with an API	Private proxy on your own servers
Where it runs	On-premises or VPC; cloud marketplaces; or Limina’s SaaS	On your own servers
How your code uses it	De-identify, call the model, re-identify	Change base_url in your OpenAI client
Putting real values back	Separate call to the re-identify route	Automatic in the response
Streaming	Not documented	Real values back in the answer
Published pricing	Free tier of 75 calls a day; paid tiers on request	Per server, with your own OpenAI key
Identifiers	Over 50 entity types	National documents and numbers for each country it serves

The practical question is how many places in your code call a model. One service you control can take three calls. Several services, scripts and a contractor’s tool are less work to point at a new address. Our overview of PII redaction software groups the options by this integration shape.

How Nonimo Gateway does it

Nonimo Gateway runs on your own servers and exposes an OpenAI-compatible API, so the integration is a change of address:

import os
from openai import OpenAI

client = OpenAI(
    base_url="https://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 puts the real values back into the response. You keep your own OpenAI key. If you route traffic through LiteLLM, there is a plugin; see LiteLLM PII guardrails.

It recognises the national documents and numbers of each country it serves: the Spanish DNI, NIE and NIF, Spain’s CUPS energy supply number, the German Steuer-ID, the French NIR, the Italian codice fiscale, the Brazilian CPF, and the Mexican RFC and CURP.

If you are still deciding whether this needs solving, start with whether putting client data in ChatGPT is a data breach. Comparing more vendors? See the Protecto alternative and the Microsoft Presidio alternative.

If you want it on your own server, one changed setting in your code and real values back in the answer, Nonimo Gateway is built for that. See plans and get in touch on the Gateway page.

Sources

Checked 20 October 2026.

• Limina, Private AI rebrands as Limina. The rename on 5 March 2026, no effect on products or customers.
• Limina, press page. The rename listed under 5 March 2026.
• Limina, homepage. 52 languages, modalities.
• Limina, pricing. The three tiers, the free 75 calls a day, file limits, deployment options and marketplaces.
• Limina Docs, prerequisites and system requirements. The CPU and GPU hardware minimums.
• Limina Docs, re-identify text. The POST /process/text/reidentify route for LLM responses.
• Limina Docs, LLM prompt redaction. The de-identify, send, re-identify pattern.
• Limina Docs, index. No proxy page.
• Limina Docs, languages. 14 core and 39 extended languages.
• Limina Docs, supported entity types. Over 50 entity types.

---

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