Guides
Practical writing about running open-source AI yourself: what the hardware actually delivers, what the licences actually say, and what the bill actually looks like.
AI assistantsOpenClaw setup: what you need before you startWhat OpenClaw actually needs to run: the gateway, a model key, chat bridges and somewhere always-on to h…
AutomationHow to self-host n8n (and what the licence actually says)A practical guide to running n8n yourself: what it needs in production, the webhook URL trap, and what t…
Private LLMRunning an LLM on CPU: what actually worksWhat a quantized model really delivers on CPU cores: tokens per second, memory needed per model size, wh…
ArchitectureA self-hosted AI stack that actually holds togetherChat, gateway, model, retrieval and observability: which open-source pieces fit together, what each one …
CostCutting an LLM bill without cutting qualityWhere model spend actually goes and the seven changes that reduce it: routing, caching, context discipli…
MCPRemote MCP server: how to host one your agents can reachMCP servers are built for local stdio. How to turn one into a remote MCP server over streamable HTTP, se…
Team chatPrivate ChatGPT for your team, on a server you controlA ChatGPT-style workspace for your team on your own domain: Open WebUI or LibreChat, your model keys, yo…
AgentsAI agent hosting: where an agent should actually runAgents need to stay up, keep state, run on schedules and hold credentials safely. Why serverless fits ba…Where to start
| If your question is | Read |
|---|---|
| Can I run a model without a GPU at all? | Running an LLM on CPU: what actually works |
| Which apps should I run together, and in what order? | A self-hosted AI stack that actually holds together |
| My API bill is growing faster than usage | Cutting an LLM bill without cutting quality |
| My agents need tools on a server, not on my laptop | How to host an MCP server your agents can reach |
| I want an always-on personal assistant | OpenClaw setup: what you need before you start |
| Can I self-host n8n, and can I charge for it? | How to self-host n8n (and what the licence actually says) |
What these guides assume
They are written for someone who can follow a README but does not want to become an infrastructure team: a developer adding an AI feature, an agency running tools for clients, an operations lead who has been asked to keep company data off third-party chat services. Numbers are from real CPU instances rather than benchmarks on hardware you will not rent, and every guide says when the honest answer is a GPU or a hosted API instead.
Each guide links to the applications it mentions, so you can go from the explanation to a running instance, and to the comparisons where a choice between tools comes up.
Put your AI stack on your own box
Pick an app, pick a size, and have it running today. Month to month, cancel whenever.