Hitch - Hermes Agent Share - Automobile Buying Assistant

Long story short. Looking for a new/used car for the wife. Visor.vin has been very helpful in doing that with my last couple of purchases. Jumped on their site and noticed it had an API. Set up an account, they gave me $1 credit to start with. The entire project cost me a buck fifty in API fees.

Set up a new agent in Hermes, pointed it to all the Visor API docs, and had it ingest and understand all those. Then gave it the API key and told it what I was looking for and got this...



While this was good for me, it is not something I can sit down with the wife with and move forward. So, I built it into a nice HTLM doc.


Much better, then I thought, the wife will want pictures, and at least one needs to be on that page, otherwise I will be clicking all night. Went to pull pics....but wait, that agent grabbing pics is shut down tight. So I set up Hermes local browser. Uses a browser on your desktop, like you would use it. Gets around a lot of barriers.

Then I started thinking about rating the cars' values. I had Hermes research and build a rubric to grade the cars on.

Then I started thinking about comparing the prices of similar models. Low and behold, Visor.vin can give that information up to two years back. I had the agent pull six months.

Then I asked the agent what else we needed and it came up with a whole host of data points that are interesting and many important.

This is where I ended up. Clickable cards that take you to the dealer site. The rubric scoring bar has popups for the data contained.



Hermes has a profile sharing system. I had the agent sanitize my files, and remove all the specifics that got built into it. The agent now is set up for any automobile.

The GitHub is here: https://github.com/frgmstr/hitch

Did not check to see if someone else had done this previously. Building it turned a slog into something fun.

One of my favorite parts of the rubric scoring...

Trim liquidity = how easily a specific trim converts into cash when it's time to sell — i.e., buyer demand that's deep enough, fast enough, and consistent enough that you never have to "pay to exit" the truck. At the listing level everyone knows used cars sell; at the trim level, liquidity is what separates a 2024 Outer Banks from a 2025 Everglades. Three ingredients:

Velocity — how fast that trim sells once listed (days on market).
Price stability — do sold examples cluster tightly around fair, or does the trim need discounts to move? Liquid trims have tight spreads between asking and sold; illiquid ones trade in noisy little-sample clusters where even a fair price is hard to defend at tax time.
Depth — how many buyers want that spec right now. High depth means your example's color/packages are somebody's dream truck instead of the leftover nobody wants.

Two practical lessons for your purchase:

Buy the liquid trims if you might resell. Outer Banks entries in our CPO section aren't just high-scoring on mileage and price — they're also the easiest to leave behind later at fair value. Wildtrak's deep MSRP discounts mask that it actually holds demand once priced right, so its rubric scores survive scrutiny.

Illiquid trims are bargains for keepers. A Heritage Edition nobody wants pays you in discount if you plan to drive it into the ground over seven years — but only a keeper should buy one, because exiting that spec is slow and ugly. If your wife's heart isn't set specifically on that look, let someone else take that risk at retail.

So when I say "trim liquidity matters," I mean: same price point can mean either easy exit or hard exit depending on which badge is on the tailgate.

That's awesome. Crazy how easy Hermes/AI has made it to get and correlate data. Just get an endpoint and off you go.

4saken said:

That's awesome. Crazy how easy Hermes/AI has made it to get and correlate data. Just get an endpoint and off you go.

Click to expand...

Working on five features to include now.

1. Standing watch with new-listing alerts. Hitch scores a report when you ask; nothing watches your saved criteria and pings you when a clean 2025 Badlands under 41k drops into the corridor or its price is cut. product-price-monitor guards one concrete item, not a criteria search across sources with dedupe — that's the whole Saturday-saving feature we're missing. The seen-list + sold/removed bookkeeping pattern from usa-car-search is directly transplantable, and Visor's days_on_market/sold semantics do most of the work without scraping seven sites.
2. Per-VIN recall status. The rubric grades recall counts by model year with web fallback. Auto.dev (or NHTSA vPIC) answers "are there unresolved recalls on this exact truck, has the dealer cleared them" — a pre-inspection red flag we currently can only check manually per VIN.
3. OEM build data. We decode trim naming from nameplates and VIN; factory-options/build sheets would settle "is this loaded example actually worth its CPO premium vs. the stripper next to it" with evidence instead of listing prose.
4. Payment/lease/OTD math. Reports flag OTD artifacts but Hitch has no calculator: payment estimate, lease money-factor math, tax+title+reg by Texas/OK zip, APR prequal band, TCO per year. Nothing public does this well — it'd be a small clean skill on auto.dev endpoints or pure arithmetic once you give me the lease terms from a dealer quote (fine-print parsing is where a lease skill would actually earn its keep).
5. Plate→VIN lookup. Minor: when a listing shows only plates, or for checking a car parked at the lot. Cheap to add alongside #2.
6. For-sale-by-owner supply. Visor + CarGurus = dealer lots. The craigslist/FB Marketplace/eBay private-party channel is where no-CPO-premium cars live — usable price evidence if you ever want to weigh "plain used from a stranger" against CPO properly.

Does this thing work for guitars?

uOpt said:

Does this thing work for guitars?

Click to expand...

No. But I am sure you could build one.