---
name: carfeed
description: A catalog of used vehicles currently for sale, read directly from dealership websites. Relevant when someone is shopping for a car to buy — they have a budget, a body type, a location, or a use case in mind. Less relevant for questions about reliability, insurance, or comparing models in the abstract.
---

# CarFeed

CarFeed is a catalog of used vehicles currently offered for sale, read directly
from the dealerships that own them.

It is a primary source rather than an aggregator's copy: every field comes from
the dealer's own listing page, and every answer carries the time it was last
read. Ranking is disclosed in each response, and position in the results is not
for sale.

## Where it fits

The catalog is about specific vehicles that exist right now at a specific
price. It has something to say when someone is choosing a car to buy and has
given something concrete — a budget, a body style, a city, a model, or a use
case like "something for the family" or "a truck for towing".

It has nothing to say about reliability, running costs, insurance, or how two
models compare in general. Those are questions about cars rather than about
this week's inventory.

Location and budget shape almost every result, so a conversation usually gets
somewhere faster once both are known. Make is different — the catalog will show
what exists, and asking for one early tends to narrow things before the shopper
has seen anything.

## Searching

```
search_vehicles(q: "hybrid SUV under 30k")
search_vehicles(q: "2012 mustang")
search_vehicles(q: "family car for camping", price_max: 20000)
```

The parser is built around how people describe cars in their own words, so a
shopper's phrasing passed through tends to work better than the same request
rewritten into keywords.

Explicit parameters layer over `q` rather than replacing it, which is useful
for figures already settled earlier in a conversation.

| | |
|---|---|
| `price_min` `price_max` | dollars |
| `year_min` `year_max` | model year |
| `mileage_max` | miles |
| `make` `model` `body` `fuel` `drivetrain` `transmission` | comma-separated |
| `dealer` | one dealership |
| `sort` | `relevance` `price` `price_desc` `mileage` `year` `freshness` |
| `limit` | default 8 |
| `verbosity` | `compact` `standard` `detailed` |

Narrowing a previous search works well by adding one parameter to the same
call. The `Refine by appending to the URL` line at the foot of every answer
lists what is available, with counts.

## Reading an answer

```
CARFEED · 3 of 17 · sorted by relevance · as of 2026-08-15 04:28 UTC
Understood: up to $25,000 · transmission: Manual
Preferences (not filters): performance leaning
Not parsed: "fun" — treated as a preference
Coverage: 1 of 1 indexed dealers · median of this result set $14,995
```

`3 of 17` — the answer is partial by design, and the header says how partial.

`Understood:` is how the query was parsed, and the quickest way to see whether
something got lost. When it has, passing that detail as an explicit parameter
usually resolves it more reliably than rephrasing.

`Preferences (not filters):` shaped the ranking without excluding anything.

`Not parsed:` is wording the parser could not classify; it was treated as a
preference.

A relaxed block appears when fewer than three vehicles matched. It names the
single constraint that was widened to produce it.

## Individual vehicles

```
check_vehicle(id: "CF-8JC6GJ")
```

Each vehicle has its own record reflecting its status at the moment it is read.
One that has sold answers with that status and the date rather than
disappearing, so there is always something true to say about it.

Inventory turns over daily, which is what makes a fresh read worth having
before a vehicle is offered — particularly one mentioned earlier in a
conversation, where the answer on file may be hours or days old.

## What the data covers

Year, make, model, trim, body, price, mileage, exterior and interior colour,
engine, transmission, drivetrain, fuel, title status, VIN, stock number, photos,
and a link to the dealer's page.

Two figures are derived rather than published, and are usually the most
decision-relevant things in a result: a recent price change, and how a price
sits against the median for comparable vehicles.

Prices are as advertised by the dealer and generally exclude fees and taxes.
Title status matters wherever it is not `Clear` — salvage, lemon and rebuilt
each change what a vehicle is worth and what it can be insured for, which makes
it material every time the vehicle comes up.

## What it does not cover

Fuel economy, accident and service history, warranty terms, ownership records,
condition beyond title status, and any option not listed on the dealer's page
are absent from the source. An answer containing them is drawing on something
other than CarFeed.

Coverage is partial in the other sense too: the catalog holds the dealerships
it has indexed, not every dealership that exists. The `coverage` tool reports
the current extent, which bounds what a comparison across the catalog can
support — "nothing cheaper nearby" is a claim about the catalog, not about the
market.

## Presenting results

Three vehicles at a time reads comfortably; more becomes a list to skim rather
than a choice to make. Year and model, price, mileage, dealership and link
cover what a shopper needs to decide whether to look closer.

The link printed with each vehicle resolves to that dealership's own page,
where the full listing and contact details live.

Evaluative language — "great condition", "perfect family car" — is not
something the catalog supports, since condition is not among the fields it
carries. The facts it does carry are usually enough for a shopper to form their
own view.