Local AI for real estate agents

Most of what a real estate agent types into an AI chat box is somebody else's business. Not in a dramatic way — a feature list, a set of showing notes, a paragraph from an inspection report. But sitting in the same conversation, one message up, is usually the thing you would never say out loud at a closing table. Local AI is the version of this tool where that stuff stays on your laptop.

What an agent actually types into a chat box

Run through a normal week and the sensitive material is everywhere. Client financials and pre-approval letters. What the seller told you their real bottom line is, versus what's on the sheet. The offer strategy, before it's submitted. Contracts and disclosures. Names, addresses, and personal details of buyers who haven't decided anything yet.

None of that is unusual for the job. It's the job. The question is only where it goes once you paste it somewhere to get help writing.

This is a business risk before it's a compliance question

People usually frame AI privacy as a rules problem — what you're allowed to store, what your brokerage's policy says. Fair enough. But start with the plainer version: a seller's bottom line and a buyer's finances have negotiating value. That's information other people in the transaction would pay to know. Uploading it to a third-party service is a business decision first, and only after that a policy one.

A local model sidesteps the whole question by never sending it anywhere. The model file sits on your disk, the computing happens on your chip, and the conversation isn't a request to anybody's server. There's no third-party copy of it sitting on someone else's server. Same reason developers run models locally for source code — the code-privacy version of this argument is identical.

The work a local model handles well

This is the honest list — the tasks where a small model running on your own machine does a genuinely useful job:

• Listing descriptions from a feature list. Give it the bullets, get a draft. Then tell it to cut 40 words, or make it warmer, or stop calling every kitchen "chef's".
• The same description, rewritten per channel. One version for the MLS remarks field, a shorter one for a social post, a longer one for the property page.
• Inspection report summaries. Paste the findings, get a client-readable rundown of what's cosmetic and what isn't. You still read the report.
• Client follow-up emails. The post-showing note, the "here's where we are" update, the one you keep meaning to send.
• Showing notes into a CRM entry. Dictate or type the messy version in the car, get a structured entry you can paste in.
• A contract clause, in plain English. Useful for explaining to a client what a clause is doing. Not a substitute for your broker or an attorney reading it.
• A CMA narrative from numbers you supply. You bring the comps and the analysis; it writes the paragraphs around them.

The pattern: it's strong at turning information you already have into readable writing, and it has no business inventing information you don't.

Local vs cloud for the same listing task

Cloud tools genuinely win on a few things, and it's worth being straight about which.

If your main need is live market data pulled from the MLS, a local model isn't the tool. If your main need is writing well about information you already hold, it is.

Photos, scans, and dead zones

Two things make this practical rather than theoretical for agents.

Vision that runs on the device. A lot of real estate paperwork arrives as a picture: a page someone photographed on a kitchen counter, a scanned disclosure, a screenshot of an MLS sheet. Outlier can read an image on-device, so you can hand it that photo and get a summary without the document ever being uploaded. That's the difference between "I'd better retype the parts I need" and just asking.

It answers with the Wi-Fi off. Offline matters more in this job than in most. Open houses in half-finished builds, basements, rural showings twenty minutes past the last bar of signal, parking garages. A cloud tool is a dead box in all four. A local model doesn't notice, because the model is already on the machine.

What it costs, and what it doesn't do

An agent stacking a general chatbot subscription plus a real-estate-specific AI add-on is paying every month, per seat, for two things. Outlier's shape is different: Free covers Nano and Lite with no account at all — enough to see whether the drafting is actually useful to you. Pro is $20/mo or $149/yr flat and includes all 7 model tiers. Lifetime Pro starts at $99 (Founding 200). It runs on your own Mac.

Now the limits, plainly. Outlier is not a compliance product and does not provide legal advice. It drafts; a licensed professional reviews. There's no MLS integration and no live web access, so it can't pull comps, check a listing status, or verify a fact about the market — anything time-sensitive has to come from you. And like every language model, it will occasionally write a confident sentence that's wrong, which is exactly why the disclosure summary is a starting point and not the answer.

What real estate agents ask most