Trial
- 50 verifications
- Chat and dashboard included
- No card required
Your AI answers. LIA Engine certifies what can be verified, and tells you what can't. No model switch, no new workflow.
Compared with
If your system already calls an AI from code, LIA Engine sits in the middle: same API, your own AI and your own key. Every answer comes back with its certificate.
from openai import OpenAI
client = OpenAI(
base_url="https://liaengine.com/v1", # before: your AI's URL
api_key="lia_...", # your LIA key
default_headers={
"x-lia-proveedor": "openai", # your AI (gemini, anthropic…)
"x-upstream-api-key": "sk-...", # your key for that AI
},
)
r = client.chat.completions.create(model="gpt-4o-mini", messages=[...])
# r.liaCertification → CERTIFIED, or what couldn't be verified
Get my LIA key
Works the same with JavaScript, curl or any OpenAI-compatible library, and with prompts in any language.
Same engine, on one screen. Upload an Excel sheet, a PDF or ask a question, and see what LIA certifies and what can't be verified.
The chat and the dashboard are in Spanish for now; your questions can be in English.
Every answer travels with its signed certificate. Your team, your client or your auditor can check it with LIA Engine's public key, without taking our word for it. If a single value in the answer changes, verification fails.
$ node verificar_certificado.js respuesta.json https://liaengine.com/v1/clave-publica OK Valid Ed25519 signature OK Answer unchanged since it was certified RESULT: VALID CERTIFICATEDownload the verifier (MIT)
You use your own AI and your own key: you pay for verification, not for someone else's tokens. Every plan includes every feature: chat, API, certificates, spreadsheets, photos and diagrams.
Payments are processed in Argentine pesos through Mercado Pago.
A service that sits between your systems and the AI you already use. Your AI answers; LIA checks the answer against your data and the rules of the request, fixes what can be computed and returns a signed certificate. Anything that can't be verified is not certified, and you're told why.
Yes. It works with OpenAI, Claude, Gemini, Groq, DeepSeek, Kimi, GLM and more, through the OpenAI-compatible API: change the base URL to https://liaengine.com/v1 and add your LIA key. Every customer uses their own AI key.
Invoices and VAT, Excel and PDF spreadsheets, photos of invoices and receipts, data extracted from documents, diagrams and charts, and electronic circuits.
If the error can be computed, LIA fixes it before returning the answer. If it can't be verified, LIA stops it: no certificate, and an explanation of why.
Every certificate is digitally signed (Ed25519). Anyone can verify it with LIA's public, MIT-licensed verifier, without creating an account.
For each request we keep the control record (date, result, model, timings), not the text of your question or the data you send. Certified answers are kept for 90 days so you can download their certificate. Details are in our privacy policy (in Spanish).
Sign in with Google, no card, and start with the free trial.
Sign-in and the dashboard are in Spanish for now. Look for «Ingresá con Google».
Sign in with Google