Skip to main content
Snaga
Try

Language-model engine

Run language models through an encrypted channel.

Aspida is an open-source Ada/SPARK engine for running and training language models. Use its client–server channel or embed the engine directly in an application.

In developmentsource build with Alire/GNAT; APIs may change during development

The Плямки of Snaga

Spark

Works with the files and tools on your computer.

Your projectChanges you can check

  • Encrypted requests

    The client encrypts requests before sending them. The trusted inference server decrypts them to run the model.

  • Models in GGUF format

    The engine reads GGUF models and includes backends for Llama, Qwen and Gemma, with optional CUDA acceleration.

  • Direct application integration

    A C-compatible library embeds the engine in a host application. A local proxy supports OpenAI-compatible clients.

Development and trust boundary

Aspida is in active development and its APIs may change. The inference server sees decrypted requests; the local OpenAI-compatible proxy also has a plaintext loopback hop on the same machine.

Source code on GitHubResearch