Installing Kompact AI

Kompact AI SBox delivers air-gapped sovereign AI inference running entirely on your existing CPU infrastructure. Installation requires zero external cloud connections or GPU drivers.

System Requirements

  • CPU Architecturex86_64 / ARM64

    Intel Xeon, AMD EPYC, or ARM Neoverse CPUs with AVX2/AVX-512 support.

  • Memory16GB+ RAM

    Minimum 16GB DDR4/DDR5 system RAM (ECC recommended for sovereign deployments).

INSTALLATION COMMAND 
# Download & Run Kompact AI SBox Installer
curl -fsSL https://get.kompact.ai/sbox-install.sh | sh

# Verify installation status
kai-sbox status --verbose

Quickstart

Get up and running with Kompact AI in under 2 minutes. Learn how to launch your first local model, send completion queries, and scale CPU threads.

Step 1: Pull a Model

Pull any domain-specific model directly into your local RAM cache:

QUICKSTART CLI 
# Pull and serve model
kai run sbox-llama3:8b --port 8000

# Query local endpoint
curl http://localhost:8000/api/v1/chat/completions \
  -H "Content-Type: application/json" \
  -d '{
    "model": "sbox-llama3:8b",
    "messages": [{"role": "user", "content": "Hello Kompact AI!"}]
  }'

Authentication & Identity

Kompact AI supports air-gapped JWT token authentication, mTLS certificate validation, and local LDAP/Active Directory identity integration for strict zero-trust environments.

HEADER AUTHENTICATION 
# Pass Bearer Token or X-API-Key header
curl http://localhost:8000/api/v1/chat/completions \
  -H "Authorization: Bearer kai_sec_789456123" \
  -H "Content-Type: application/json"

Cluster & node Mgmt

Manage multi-node CPU clusters, inspect NUMA node memory balancing, and control thread allocation per core dynamically.

CLUSTER STATUS 
# Inspect active nodes and CPU NUMA topology
kai cluster nodes list
kai cluster numa-balance --auto

Model Registry

Inspect all registered models across SBox, RBox, EBox, FinBox, and HBox. Quantize, version, and air-gap custom GGUF/Safetensors checkpoints.

REGISTRY COMMANDS 
# List available offline models
kai models list

# Register custom model weights
kai models register --name custom-llm --path ./models/weights.gguf

Chat Completions

POST /api/v1/chat/completions

Quick start guide

The Chat Completions API allows you to send a sequence of messages and receive a model-generated response. It is fully compatible with OpenAI REST specifications, enabling seamless zero-code-change migrations for existing applications.

Requests are executed directly on your sovereign Kompact AI hardware cluster (SBox, EBox, RBox, FinBox, HBox) with private, zero-data-leakage inference.

Request Body Parameters

  • model string required

    ID of the model to use (e.g. kai-sbox, llama-3-8b, qwen-2.5).

  • messages array required

    A list of messages comprising the conversation history. Each message object requires a role (system, user, assistant) and content.

  • temperature number optional

    Sampling temperature between 0 and 2. Higher values like 0.8 make output more random, lower values like 0.2 make it more deterministic. Default is 0.7.

  • max_tokens integer optional

    The maximum number of tokens to generate in the completion.

REQUEST EXAMPLE

Health

GET /api/v1/health

Health Check Overview

The Health endpoint provides real-time diagnostic status of the Kompact AI local inference engine, active model instances, NUMA node memory utilization, and CPU thread health.

Use this endpoint for automated liveness/readiness probes in Docker, Kubernetes, or custom sovereign cluster orchestrators.

Response Attributes

  • status string

    Overall engine health: "ok", "degraded", or "unhealthy".

  • uptime_seconds number

    Elapsed execution time of the active KAI daemon in seconds.

  • active_models integer

    Number of LLMs currently loaded in system RAM / NUMA nodes.

REQUEST EXAMPLE