Blog
No hype, only what's real.
How it works and why it was built this way. Each post is long-form with a summary at the top, so you can read the whole argument or just the claim.
How it works
Short films, one noun at a time: what a Host is, the machine it runs on, the network it lives on, and where a request goes. Each post is the film plus the claims it draws, written down.
· updated · Part 1 · Film
What a Host is: a machine of your own on your network
A machine of your own on your network: one pod, three containers, its own disk and database, two addresses and no public one, a certificate, one outbound line.
· updated · Part 2 · Film
Your machine
Inside the pod: three containers, two disks and what is on each, where the certificate lives, and the rooms beside it that exist only while you use them.
· updated · Part 3 · Film
The network you own
Who may join a Host's network, how a home machine reaches a pod closed to the internet, which way traffic leaves and under whose name, three lanes, Host to Host.
· updated · Part 4 · Film
How the Host handles a request
Four kinds of agent on a Host, why the Host is not one, and where one request goes: a deterministic call, a scoped agent, or a coding harness.
· updated · Part 5 · Film
What it remembers
A row per turn, threads instead of sessions, entries derived on a schedule and superseded rather than overwritten, and a search that names its own gaps.
· updated · Part 6 · Film
What it cannot do
Five checks in code: a zero floor for agents, a gate at the operation, a refusal at the door, a second factor to widen, and no code outside the Host's own door.
· updated · Part 7 · Film
Building with it: seven coding agents on your own keys
The Workbench: Claude Code, Codex, Cursor and four more coding agents on your own keys, a plan you approve first, and one rolling pull request you merge.
· updated · Part 8 · Film
Data that stays home: a folder indexed on your own machine
Your computer joins the network by running one program. It keeps its folders, answers questions about them on your hardware, and can serve the model itself.
· updated · Part 9 · Film
The Distributed Context Network
Your files stay where they are. The Host holds a route map — what exists, where it lives, who may touch it — and every read travels to the data.
Essays
Longer arguments about how this kind of system should be built — the reasoning, and what it rules out.
· updated
Dots, Muse, Grok Bot or OpenClaw: who holds your AI agent's memory?
Always-on AI agents come in two kinds: the model company runs it, or you run it yourself. What each asks you to trust.
· updated · Film
One server, six Hosts, one new hire
A team of six on one server. How context is granted to a Host instead of a drive: indexed in place, sanitized, scoped to a role, and set to expire.
· updated · Film
What is Privatae?
A network you own, with a machine on it that is yours. What the Host does, what the network around it is, and what we are building it for.
· updated · Film
You never chatted with an agent
Agents ran on computers for decades before AI: scoped, exact, repeatable. What the word means now, what it meant first, and how agents work on a Host.
· updated
AI agent containment is a networking problem
Almost everything an AI agent does is a network operation. The controls that held through two years of agent breaches were network controls.
Subscribe: RSS.
Posts from before 19 October 2026 describe Hosts on the Privatae platform, which is no longer offered to new customers. What Privatae offers now: pricing.