WORKLOADS - ENCRYPTED COMPUTE
Production-ready FHE applications, plus an SDK to build your own.
Encrypted workloads are complete applications that perform computations on encrypted data using Fully Homomorphic Encryption (FHE). Unlike traditional systems that require decryption before processing, these workloads maintain data privacy throughout the entire computation pipeline from input to output.
We provide both a catalog of pre-built, production-ready workloads for common use cases, and a developer library that enables you to build custom encrypted applications without deep FHE expertise. All workloads run on our GPU-accelerated infrastructure, allowing you to test and deploy at scale without managing local hardware.
CATALOG
Production-ready encrypted compute workloads. Explore interactive demos or contact us for access.
Sketch a digit, then run the real FHE pipeline on our live demo: keys generated in your browser, ciphertext sent to the server, prediction decrypted client-side.
Run the real encrypted inference flow directly from this page.
Run convolutional neural network inference on encrypted image data. Process visual inputs while keeping your images completely private throughout the computation.
Explore the CNN inference workload details and live experience.
Open Demo →Analyze and classify text on encrypted data. Perform sentiment analysis, topic classification, and other NLP tasks while keeping your text completely private.
Apply filters and transformations to encrypted images. Enhance, sharpen, or process visual data while maintaining complete privacy throughout the operation.
Generate predictions and insights from encrypted data. Build risk models, forecast outcomes, and analyze trends while keeping source data completely private.
Perform similarity search and exact match queries on encrypted vector embeddings. Ideal for semantic search, RAG applications, and recommendation systems.
Experience encrypted vector search with sample embeddings.
Contact UsCompute statistical aggregations and analytics on encrypted datasets. Calculate sums, means, correlations, and other metrics without decrypting source data.
Detect fraudulent patterns and anomalies in encrypted transaction data. Identify suspicious activities while maintaining complete privacy of financial records.
See fraud detection on encrypted transaction data.
Contact UsGenerate personalized recommendations from encrypted user data. Build collaborative filtering and content-based systems while protecting user privacy.
Generate recommendations from encrypted user data.
Contact UsDEVELOPER LIBRARY
Our developer library abstracts away FHE complexity, letting you build encrypted applications using familiar programming patterns.
The library provides a high-level API that handles FHE operations, parameter selection, and optimization automatically. You write code using familiar tensor operations and mathematical functions, and the library translates these to efficient encrypted computations under the hood.
from lattica.deployer.hom_ops import SequentialHomOp, HomLinear, HomSquare from lattica.deployer.client_ops import ClientReshape from lattica.deployer.admin import admin_api hom_mnist = SequentialHomOp( # preprocess: flatten 28x28 input image to a 784-d vector ClientReshape((BATCH_SIZE, 28 * 28,)), # first linear layer HomLinear(l1_weight.shape), # square activation HomSquare(), # second linear layer HomLinear(l2_weight.shape), # postprocess: reshape output to 10-class vector ClientReshape((BATCH_SIZE, 10,)), ) admin_api.deploy(hom_mnist, n=2**14, q_bits=124, device='gpu-L4-aws')
Use standard tensor operations and mathematical functions. No need to understand FHE schemes, parameter selection, or cryptographic primitives.
Test and deploy workloads directly on our GPU infrastructure. No local setup required; develop, test, and scale in the cloud.
The library automatically selects FHE parameters, handles noise management, and optimizes computation graphs for performance.
Built on HEAL, your workloads automatically benefit from the latest GPU hardware without code changes. Write once, optimize everywhere.