An AI agent that runs on hardware you control, so your prompts, files and conversations stay where you decide. You know what it can do, who can talk to it and how to switch it off.

One supported install target: a Mac, an Ubuntu LTS server or one cloud VM you own

One agent platform: OpenClaw by default. I've built an integration with it and tested it against a live OpenClaw instance. Another open-source agent platform only if we confirm it works on your machine before you pay.

One model-provider connection: a hosted model on your own API account, or a local model through Ollama on hardware that can run it

One channel or integration

Basic access controls: an allowlist of users, least-privilege credentials and limits on what it may do without asking

Start/stop instructions and logging

Install verification against a written checklist

A runbook and one handover session (up to 45 minutes)

14 days after acceptance to fix anything that fails the agreed checklist

Enterprise hardening or security certification, any guarantee of security, GPU clusters, complex networking (VPNs, VPC peering), high availability, large migrations, 24/7 monitoring, and hardware, infrastructure, model or API fees. You pay those directly, on your own accounts.

5 working days from kickoff. Kickoff is when I have admin access to the machine and the channel token.

The machine, server or cloud VM it will run on, with admin access

The channel account or bot token, created by you

A short list of what the agent should and shouldn't do

Your model choice, or a budget and privacy requirement so I can recommend one

Anything needed to pass the agreed checklist is fixed at no charge. Another channel, platform or machine is a change request, quoted in writing first.

The agent starts, answers on the agreed channel and stops cleanly on the agreed machine

The install checklist passes, and you have a copy

The configuration is written down in the runbook

We've held the handover session

I tell you before you pay. A second channel is the extra channel add-on ($250). Local models for several users and search over your documents is the Private AI Environment (from $1,250); a stack you choose with TLS and up to three integrations is Custom Deployment (from $2,000). If your machine can't run the model you want, I say so and suggest what can. Complex networking (VPNs, VPC peering) or a second machine gets a written quote. If something unexpected turns up after kickoff, such as an upstream bug or a hardware limit, I stop and give you the options. The price changes only if you accept a written change.

Private AI Environment: from $1,250. Local models through Ollama, sized to your hardware, an open-source chat interface, chosen in scoping, for up to 10 named users, and search over one folder of your documents.

Custom Deployment: from $2,000. An open-source AI stack you choose, deployed with Docker Compose, TLS, up to three integrations and an update and rollback procedure.

AI Infrastructure Engineering: from $3,000. Model serving, a document-search pipeline (keyword plus embedding search, with source citations) and an agent runtime on one or two hosts. GPU setup only on hardware we confirm in scoping; I don't take on GPU clusters.

Maintenance: from $200/month. Updates, a monthly check and small fixes.

Extra channel $250 · Extra integration (documented API) $300 · Deployment help for a non-standard host $300 · Extra handover session $125 · Priority delivery +30% (when my calendar allows)

Evidence-Backed AI Operations Console (Prototype: new screens on the data structures of a private, local-first evidence system I built in Postgres, which records events from agent runtimes including OpenClaw; data shown is illustrative)

Message me on Contra with where it should run (machine and operating system), the channel you want and what the agent should do. I'll reply within two working days and confirm whether it fits the $600 setup.

Cover: an illustrative interface I designed for this service, not client work. Status: Concept. The screen shows Stillroom, a private AI node for Hollin & Vey (demo), an invented law office. It shows outbound traffic, memory use and what the agent may do, in three words: Allowed, Ask first, Blocked. The host, traffic, model labels and permission list are invented. I have not set up a node for anyone. The local models, chat interface and document index on the screen belong to the Private AI Environment, from $1,250.