Insights · Research · Field Notes

Enterprise AI, explained for builders and buyers. Page 23 of 43

Practical writing on governed agent orchestration, on-premise AI, compliance, and the infrastructure decisions that separate pilot projects from production platforms.

386 Articles
43 Topics
12 Featured
Padlock resting on a computer keyboard, representing the security and provenance review that comes before self-hosting DeepSeek open weights on premises
AI Security 11 min read

DeepSeek On-Premise: A Provenance and Risk Review Before You Self-Host V4

Self-hosting DeepSeek weights is a different risk decision from using DeepSeek's app or API. This review covers the September 2026 lineup, what public-sector restrictions actually cover, a provenance checklist for the weights, and the GPU arithmetic.

#DeepSeek#open-weight models#local LLM
Read article
Close-up of a memory module with DRAM chips and gold edge contacts, a reminder that memory capacity and bandwidth decide which local models a desk-side AI system can run
AI Infrastructure 11 min read

DGX Spark vs Mac Studio vs Ryzen AI Max+ 395: Local LLM Hardware Compared (2026)

NVIDIA DGX Spark, Apple's M5 Max and M5 Ultra Mac Studio, and AMD's Ryzen AI Max+ 395 systems compared from their spec sheets: memory, bandwidth, stated compute, software stack, clustering, power and price, and where each fits in an enterprise AI plan.

#local AI hardware#DGX Spark#Mac Studio
Read article
Developer in a dimly lit room reading code on two monitors, representing hands-on evaluation of open-source AI agent harnesses
AI Platforms 10 min read

Hermes Agent vs OpenClaw: Architecture, Skills, Security and Fit

A neutral comparison of the two most popular open-source personal agent harnesses: how OpenClaw's gateway and Hermes Agent's learning loop differ in architecture, skills, memory, channels, local models, sandboxing, security record and stewardship, and which suits which job.

#Hermes Agent#OpenClaw#agent harness
Read article
Rack of enterprise server equipment in a dark data hall, representing the multi-node GPU capacity a 2.8-trillion-parameter Kimi K3 deployment requires
AI Infrastructure 8 min read

Kimi K3 On-Premises: What Serving a 2.8-Trillion-Parameter Open Model Really Takes

Moonshot AI published the full Kimi K3 weights in July 2026. Self-hosting them is allowed, but the checkpoint alone is about 1.56 TB. Here is the license in plain terms, the memory arithmetic per node, what multi-node serving involves, and when a smaller model is the better call.

#Kimi K3#open-weight models#mixture of experts
Read article
Combination padlock resting on a white computer keyboard, representing the access controls an enterprise needs around a self-hosted OpenClaw gateway
AI Security 11 min read

OpenClaw Security for Enterprises: Risks, CVEs and Controls

What OpenClaw's own security model assumes, which advisories and skill-registry incidents are on the public record, how to harden a gateway using the official documentation, and when an enterprise should move a workload to a governed platform instead.

#OpenClaw#AI security#AI agent governance
Read article
An engineer adjusting components inside an illuminated server rack, representing the GPU tiers that decide which Qwen model an enterprise can run on premises
AI Infrastructure 9 min read

Qwen On-Premise: Choosing a Qwen3.5, 3.6 or 3.8 Model by Size Tier

The Qwen family now runs from sub-1B edge models to a 2.4-trillion-parameter flagship, with different licenses at the top end. A tier-by-tier guide to dense versus MoE, coder and vision variants, official FP8 and Int4 builds, long context, tool calling and the GPUs each tier needs.

#Qwen#open-weight models#local LLM
Read article
Inside a desktop PC: a motherboard with an M.2 drive installed beneath a lit Radeon graphics card, representing workstation hardware for running a local LLM
AI Infrastructure 9 min read

How to Run OpenClaw and Hermes Agent on a Local LLM

A practical setup guide for pointing OpenClaw or Hermes Agent at a model on your own hardware: choosing between Ollama, vLLM and llama.cpp, the documented configuration for each harness, Docker isolation, MCP servers, and how to size a tool-calling model to your GPU memory.

#OpenClaw#Hermes Agent#local AI infrastructure
Read article
Three colleagues reviewing work on laptops in an open-plan office, representing relationship managers preparing client briefings with a private AI assistant that respects bank information barriers
Finance AI 6 min read

AI for Relationship Managers: Client Briefings Without Breaching Information Barriers

A relationship manager assistant that pulls client context from every system is an obvious win and an obvious risk. In a bank with private-side functions, the naive design turns the retrieval index into a hole in the information barrier. How to build the assistant so the wall survives it.

#financial services AI#private AI#on-premises AI
Read article
Euro banknotes, a tablet showing a price chart, a calculator and printed financial reports on a desk, representing the capital and operating budget decisions behind on-premises AI infrastructure
Enterprise AI Economics 7 min read

CapEx or OpEx? How Finance Teams Account for On-Premises AI Spend

The same AI programme can land on the balance sheet or in the income statement depending on how it is bought and built. A line-by-line guide to how hardware, licences, internal agent development, fine-tuning and data work are typically classified under IFRS and US GAAP, and where AI creates new judgement calls.

#AI platform TCO#enterprise AI investment#on-premises AI
Read article
Stay ahead of enterprise AI

Turn insight into an on-prem AI roadmap.

Pair these articles with a product walkthrough to see how VDF AI handles orchestration, governance, and cost control inside your own infrastructure.