# OwnerLM nodes

Private AI tools for real estate work, plus GPU inference capacity installed
inside qualified properties. The live MCP server connects approved agents to
an owner's private operating context. Hosted compute turns usable power and
secure space into an income source while powering the property's own AI.
Page: https://ownerlm.com/compute/nodes

## What this is

A node is a single chassis of AMD or NVIDIA accelerators installed in a
property that already carries the power, space, and cooling. Not a data
hall, not a region code. A known street address in the metro your users
and devices are already in.

Capacity is added as qualified properties complete site review and installation.
Reservations are real and answered by a person, usually the same day. There is
no self-serve provisioning API yet.

## What is live right now

The OwnerLM MCP server. Streamable HTTP, JSON-RPC 2.0.

    https://ownerlm.com/api/public/mcp

    Accept: application/json, text/event-stream

Two ways in. Clients that support OAuth 2.1 (Claude, ChatGPT, Cursor)
self-register, then send the human through one sign-in and approval
screen; after that the agent acts as that user under their workspace
permissions. Clients without OAuth use a bearer key created in Settings,
Developer inside any OwnerLM workspace:

    Authorization: Bearer owlm_your_api_key

Around forty tools covering portfolios, properties, documents with semantic
search and citations, deals, vendor contracts, reports, background skills,
and agent runs. Full reference: https://ownerlm.com/docs/mcp and the raw
markdown at https://ownerlm.com/mcp.md

Behind the tools is a harness: posture rules on what an agent may do,
approvals for the writes that need a human, a full action ledger, and
one-click undo from before and after snapshots. We plan to open source it.

## Who this suits

Track 1, real estate and proptech AI. Agents working leases, financials,
work orders, rent rolls, and whole portfolios. Use the tools today, and put
the model that reads them on hardware near the same properties later.

Track 2, edge inference and IoT. Latency-bound work: cameras, access
control, property systems, small models kept resident next to the sensors
that feed them. Steady streaming inference rather than bursty training.

## When it is not the right tool

- You want a hosted frontier model endpoint to call. This is capacity and
  tooling, not a model API.
- You need thousands of GPUs for a training run this week. These property-based
  installations are designed for steady AI work rather than massive training runs.
- The work has no latency or residency constraint and no real estate in it.
  A large cloud region is simpler.

## How to reserve

Fill the form at https://ownerlm.com/compute/nodes with a work email, the
track, rough accelerator and count, the metro, and a timeline. Or call or
text 816.300.7645.

Each property is quoted separately. There is no public rate card because the
number follows the property's power, installation, and term.

## Related

- https://ownerlm.com/compute - the compute program and how nodes fit a property
- https://ownerlm.com/compute/market - available and planned capacity by site
- https://ownerlm.com/docs/mcp - MCP setup and tool reference
- https://ownerlm.com/llms.txt - the rest of the site, for machines