ZMS Digital Solutions

Self-hosted LLM inference with no proxy in the middle

Run quantized open models on your own CPUs, expose an OpenAI-compatible API, and bill for it — inference engine, gateway, metering, and dashboards in one platform.

Create account Sign in

from openai import OpenAI

client = OpenAI(base_url="https://your-host/v1", api_key="mf_...")
client.chat.completions.create(
    model="zms-coder-7b",
    messages=[{"role": "user", "content": "Explain mmap"}],
    stream=True,
)

OpenAI-compatible

Drop-in /v1/chat/completions with SSE streaming — point any existing SDK at your own host.

CPU-first runtime

llama.cpp with mmap-backed GGUF loading, a RAM-budgeted LRU pool, and continuous batching.

Metered by default

Per-key rate limits, monthly token quotas, and margin analytics on every request.

Isolated by design

Engine, gateway, and control plane run as separate processes; gRPC never leaves the host.