Abraham Gómez

Zyrabit SLM is an open-source project entering beta/MVP. It explores how developers and organizations can run useful AI systems close to their data, including offline or controlled environments.

See it in action: truly sovereign AI

This hands-on demo shows Zyrabit Platform running 100% offline: processing voice commands and capturing images in real time without external APIs or an internet connection.

What works today

Public repository and documentation.
Reproducible installation through the repository lifecycle scripts.
Local model inference through the documented stack.
Initial benchmark work on Apple Silicon.
Examples for retrieval, model orchestration, and operational checks.

The repository describes a Docker-managed stack, an Ollama-backed model network, a RAG API, ChromaDB, Prometheus, Grafana, MCP, and optional n8n. These are components documented by the current repository, not a claim that every deployment is production-ready.

The problem it explores

Cloud APIs are useful, but they are not the only deployment shape. Some teams need local execution, inspectable workflows, and the ability to keep data within an environment they control. Zyrabit explores what that layer could look like with hardware that is already available.

A measured example

The Mac case study compares DeepSeek-R1-Distill-Llama-8B-Q4_K_M.gguf and qwen2.5-7b-instruct.gguf on an Apple M1 Pro with 16 GB of unified memory. The same question produced different latency and memory trade-offs. Read the full measurement.

What this does not claim

Zyrabit is not presented as certified infrastructure for every regulated organization, a medical diagnostic system, a universal benchmark, or a zero-cost production platform. Those claims require separate evidence.

Links

The project is available at Zyrabit-tech/zyrabit-SLM. The repository and its current README are the source of truth for installation and supported capabilities.