The private computer for AI.

Build a chat your users can trust with anything, because their data stays sealed. Run agents that leave a permanent, verifiable trail of every action.

Try the demo Integrate the SDK

Live on testnet
Core audited
Mainnet next

run-7f3a.recordAttested

app      support-agent
model    small open model
where    guardian g-03 · Intel TDX

## run
prompt   encrypted
files    3 · read inside the TEE only
answer   412 tokens · readable only with your key

## run
01 input   encrypted
02 decrypt inside the TEE only
03 infer   ✓ done
04 output  encrypted

## attestation ✓ Attested

hash 0x9be1…4c07on-chain

Everyone wants AI. Privacy is holding it back.

Your data is what makes AI useful. Today, users hand it over to whoever runs the model.

53%

of IT leaders say data privacy is their top concern with AI agents.

Cloudera, 2025

27%

of companies have banned generative AI over privacy and data risks.

Cisco, 2024

46%

of people are willing to trust AI. The rest hold back.

KPMG and University of Melbourne, 2025

How it works

Run the same model. Show no one.

The model runs inside secure hardware on a guardian node. Only that hardware can read your data, and it signs every answer so you can check what ran.

Private inference

Prompts and outputs are decrypted only inside secure hardware.

Private data

Your data is used to answer and stays yours.

A receipt for every run

What ran and where, attested and anchored on-chain.

Private models Coming next

Serve your own model without exposing its weights.

Accountable Privacy

A receipt for every run.

If agents act like people, they should answer to the same rules of ethics and legality. Palliora AI records every run end to end, and no single party can hold, edit or quietly delete that record.

Private by default

Your key opens your records.

No single party decides

A record opens only when most independent guardians agree.

Every request on the record

Every request is logged on-chain.

Platform-held logs

One company keeps the log and can change it.

Palliora AI record

Sealed on-chain. Only a guardian majority can open it.

Your records, your call. Only your key opens a record. Share a time-bound view key with anyone you choose, for as long as you choose.

Who it's for

Start where models are used

Now

People using models

Plug into routers, harnesses and AI apps.

Specialist model builders

Train a model for one job and serve it privately.

Then

Enterprises

Run AI on sensitive data with a record auditors can check.

Get started

Pick your way in

For builders

Try it on your own data

Run a model privately in the demo app and see the attested record.

Try the demo

For platforms

Offer private AI

Routers, harnesses, AI apps and model builders add private compute with one SDK. Pay per run in fiat or stablecoins.

Integrate the SDK

FAQ

Questions

Who sees my data?

Only the model, inside secure hardware on the guardian node. It stays encrypted everywhere else, including for the operator.

What runs today?

A small open model inside Intel TDX on testnet guardians. Larger models come as guardians with stronger hardware join.

Who can read my records?

You, and anyone you give a view key. Anyone else needs a majority of independent guardians to agree.

How do I plug it in?

Through the SDK, inside the router, harness or AI app you already use.

How do I pay?

Per run, in fiat or stablecoins.

Something we didn't cover? Ask us directly.

Run AI on your data. Show no one.

Try a private run in the demo, or add Palliora AI to your product with one SDK.

Try the demo Integrate the SDK