Build, deploy, generate keys, and query an encrypted pipeline

Lattica Studio is one workspace for the full life of a homomorphic pipeline. You build it from Lattica's operators, deploy it to our platform, generate keys once, and run queries against it.

The lifecycle

Four stages, in order. Build is the part you write. Deploy, Key generation, and Query are mostly handled for you. Open any stage to see its steps.

STAGE 1 · YOUR WORK

Build

assemble a pipeline from HomOps3 steps

on your machine · the part you write

1

Compose with HomOps

operators shaped like standard ML ops, with extra parameters for FHE

2

Set the structure

arrange your HomOps into the computation graph

3

Define parameters

shapes, weights, and encryption params

STAGE 2

Deploy

deploy_pipeline()3 steps

studio.deploy_pipeline()

On upload or compile error, fix the pipeline and deploy again from Build.

1

Register or reuse

create a model, or redeploy into the same name

2

Upload

pipeline + params

3

Compile

the backend turns it into an FHE circuit

STAGE 3

Key generation

create token, generate keys2 steps

tokens.create + generate_key

1

Create access token

studio.tokens.create, saved locally for this model name

2

Generate keys

secret key stays; evaluation key uploads and registers

STAGE 4

Query

run encrypted queries against your pipeline4 steps

query loop

1

Start worker

credits begin

2

Load existing key

generate_key(load_if_exists=True)

3

run_query()

repeat as many times as you need

×N

4

Stop worker

credits stop

Click a stage to collapse or open it. Build runs on your machine. Deploy, Key generation, and Query run on the platform.

Where it runsYour machineLattica Studio + platformToken and keysWorker

Most of your time is here

Deploy, Key generation, and Query are short because the platform does the work. Build is where you spend your time. You write the pipeline yourself, using HomOps, operators shaped like standard ML ops, with extra parameters added for FHE. A linear layer, an activation, a matmul each has a homomorphic match, though a few HomOps, like HomModSwitch and Bootstrap, exist only to maintain the ciphertext and have no ML counterpart. You arrange them into a pipeline and set the parameters: input shapes, weights, and the encryption settings. Once it runs, Deploy, Key generation, and Query take over.

Read how to build a pipeline →

What this is for

Lattica Studio is for researchers and developers who want to take a model, make it run under encryption, and test it end to end without leaving one environment.

Pipeline

Your computation, expressed in homomorphic operators.

HomOps

The homomorphic operators you build a pipeline from.

License

Authenticates you to the platform for build and deploy work.

Access tokenquery token

Scoped to one deployment. You create it after deploy, then use it to run queries against the pipeline.

Worker

The accelerator that runs your encrypted pipeline. It bills by runtime, so you start it when you query, and stop it when done.

Where to go next

Read the concepts first if homomorphic pipelines are new to you. Otherwise, start at the stage you care about.

Start here

Concepts →

The mental model: pipelines, operators, keys, and tokens, in plain terms.

Stage 1 · the main work

Build the Pipeline →

Arrange HomOps into your pipeline, then set shapes, weights, and encryption parameters.

Stages 2 and 3

Deploy the Pipeline →

Upload and compile, then create a token and generate keys.

Stage 4

Run Queries →

Start a worker, load keys, send encrypted queries, read decrypted results, stop the worker.

See it all at once

End-to-End Example →

Build, deploy, generate keys, and query in one script. Copy it, run it, see the result.