Live AI · Open source

Watch an AI call the race — before it runs.

greyhound-ai reads a racecard and returns a ranked prediction with its reasoning. Every call here is made before the off, then scored against the real result — nothing cherry-picked. Built on the GreyhoundAPI, and open source so you can run it yourself.

How often the top pick is right

48%Top pick won
76.5%Top pick placed
490Races scored

Share of races where the model's number-one pick won outright, and where it finished in the first three.

Is the confidence honest?

70%+1268 picks
said 79%
won 28%
55–70%569 picks
said 60%
won 9%
40–55%885 picks
said 45%
won 8%
Under 40%218 picks
said 29%
won 6%

Every runner grouped by the confidence the model gave it. If it's well-calibrated, "said" and "won" line up.

How close does it get?

Winner was the model's #1 pick
48%
Winner was in the model's top 2
67%
Winner was in the model's top 3
76%

How often the eventual winner appeared near the top of the ranking, over 490 scored races.

greyhound-ai · GreyhoundAPI/greyhound-ai GitHub
Browse the calls
Show
1 of 60
greyhound-insights · GreyhoundAPI/greyhound-insights GitHub

Ask the data a question in plain English — greyhound-insights fetches the real races and answers with what it found. Recent examples:

QWhat race distances have run recently at Romford, and were there any standout winners?

Recently at Romford, the following race distances were run:

  1. 380 meters
  2. 500 meters
  3. 640 meters
Standout Winners:
  • 500 meters (Race ID: 8216514):
    • Winner: Blue Falcon 2
      • Trainer: H F Keightley
      • Run Time: 30.84 seconds
      • Comment: Led 1, kept on well.
  • 640 meters (Race ID: 8216237):
    • Winner: Bawn Rebel 6
      • Trainer: S A Kimber
      • Run Time: 39.11 seconds
      • Comment: Always in command.
  • 380 meters (Race ID: 8216235):
    • Winner: Droopys Zephyr
      • Trainer: R M Yeates
      • Run Time: 30.68 seconds
      • Comment: Bossed it from traps.

These results highlight some of the recent performances at Romford, showcasing both the distances and notable winners.

Answered withlist_trackstrack_racesget_race
QHow competitive was the latest race at Nottingham — a clear favourite, or close?

The latest race at Nottingham was quite competitive. It was a 500m A1 race with a field size of 6 dogs. The winner, Storm Ghost, finished with a time of 30.77 seconds, leading from the start. The second-place dog, Rapid Icon, finished approximately 2.11 lengths behind, with a time of 30.92 seconds.

The starting prices (SP) for the top two dogs were:

  • Storm Ghost: 5/2 (3.5 in decimal)
  • Rapid Icon: 6/1 (7 in decimal)

This indicates that while Storm Ghost was the favorite, the margin of victory was not excessively large, suggesting a competitive race.

Answered withlist_trackstrack_racesget_race
QWhat were the most recent results at Sandown Park in Australia, and who won?

The most recent race at Sandown Park in Australia was Race 11, which was a Grade A1 race over a distance of 500 meters. The race was completed, and the winner was Clona Sniper 2, trained by L J Mullins, with a run time of 30.81 seconds.

Here are the top three finishers:

  1. Clona Sniper 2 - 1st place
  2. Bawn Express 4 - 2nd place, finished 1.63 lengths behind
  3. Bawn Nitro 3 - 3rd place, finished 4.22 lengths behind

The race had a total of 6 runners.

Answered withlatest_results
QLooking at recent Romford results, which trainer has had the most winners?

In the recent results from Romford, the trainer with the most winners is H F Keightley, who had 1 winner with the dog Blue Falcon 2 in race ID 8216514 on July 9, 2026.

Other trainers had the following results:

  • J R Liddle: 0 winners (best position: 2nd with Slippy Pilot 3)
  • D P Mullins: 0 winners (best position: 3rd with Newinn King)
  • A B Judge: 0 winners (best position: 4th with Newinn Hurricane 3)
  • M L Wallis: 0 winners (best position: 5th with Rathcoole Bolt 2)
  • S A Kimber: 0 winners (best position: 6th with Glenside Flyer)

Thus, H F Keightley stands out as the only trainer with a winning dog in the recent Romford results.

Answered withlatest_resultstrainer_results
QAcross recent races at Nottingham, which traps have won most often?

In recent races at Nottingham, the following traps have won the most often:

  1. Trap 1:
    • Won in Race ID 8216513 with the dog "Storm Ghost".
    • Won in Race ID 8216223 with the dog "Bawn Quest 2".
  2. Trap 2:
    • Won in Race ID 8216226 with the dog "Magic Maverick".
  3. Trap 4:
    • Won in Race ID 8216225 with the dog "Glenside Hurricane 2".
    • Won in Race ID 8216224 with the dog "Rapid Rebel".
  4. Trap 6:
    • Won in Race ID 8216222 with the dog "Magic Sniper".

From the data, Trap 1 and Trap 4 have each secured two wins, while Trap 2 and Trap 6 have one win each.

Answered withlist_trackstrack_racesget_race
QShow me the most recent completed races at Romford and who won each one.

The most recent completed race at Romford was Race 11, which took place over a distance of 500 meters. Here are the details:

  • Winner: Blue Falcon 2
    • Trainer: H F Keightley
    • Starting Price: 5/2
    • Run Time: 30.84 seconds

The second place was Slippy Pilot 3, trained by J R Liddle, finishing 1.89 lengths behind the winner. The third place was Newinn King, trained by D P Mullins, finishing 3.6 lengths behind.

Answered withlatest_results

Run it yourself

greyhound-ai is open source. Point it at any race on the API and you get the same ranked preview — preview, analyze and backtest — straight from your terminal.

View on GitHub
greyhound-ai
$ python -m greyhound_ai preview 8005458

{
  "shape": "Early speed likely from traps 1 and 2…",
  "ranking": [
    { "trap": 1, "dog": "Rising Rebel", "rank": 1, "confidence": 0.90 },
    { "trap": 2, "dog": "Slippy Voyager", "rank": 2, "confidence": 0.70 }
  ],
  "model": "gpt-4o-mini"
}
# scored against the real result — see the Accuracy tab above

These figures are a true out-of-sample backtest — each race is predicted before its result is known, then scored — but the races come from a synthetic sandbox dataset (demo cards, not real GB/AU results), so they show the pipeline works end to end, not a real-world edge. Point greyhound-ai at the live feed and the same backtest produces real numbers. It is for research and education and is not betting advice.