AioTraceMoeAPI

A simple but extensible asynchronous Python wrapper for the trace.moe API.

Key Features

• Async: Built on top of httpx.
• Typed: Fully typed (ships py.typed), with Pydantic v2 models for every response.
• Complete: Search by URL, file, bytes or ColorLayout vector (including batch search), /me quota and usage history, preview URL helpers.
• Precise errors: A dedicated exception for every API error, with optional automatic retries.

Installation

Python 3.10+ is required.

uv add aiotracemoeapi

or

pip install aiotracemoeapi

Usage

Search by URL, file or bytes

import asyncio

from aiotracemoeapi import TraceMoe


async def main():
    async with TraceMoe() as api:
        # URLs are detected automatically
        response = await api.search("https://images.plurk.com/32B15UXxymfSMwKGTObY5e.jpg")

        # A local file path, pathlib.Path, bytes or a binary file object work too
        # response = await api.search("image.jpg")
        # response = await api.search(open("image.jpg", "rb"))

    best = response.best_result
    if best is None:
        print("No results found.")
        return

    print(f"Anime: {best.anilist_info.title.romaji}")  # AniList info is included by default
    print(f"Episode: {best.episode_start}")
    print(f"Scene: {best.anime_from:.1f}s - {best.anime_to:.1f}s, best frame at {best.at:.1f}s")
    print(f"Similarity: {best.short_similarity()}")  # below 90% is most likely a wrong result
    print(f"Preview: {best.video_url(size='l', mute=True)}")  # preview URLs expire in 5 minutes
    print(f"Quota used: {response.quota_used}/{response.quota}")


if __name__ == "__main__":
    asyncio.run(main())

Search options:

from aiotracemoeapi import CutBorders

await api.search(
    "image.jpg",
    anilist_id=21034,  # search only within one anime
    anilist_info=False,  # return only AniList IDs (faster)
    cut_borders=CutBorders.BOTH,  # search with both the original and the cut image (may cost 2 credits)
)

Search by ColorLayout vector

If you already have the 33-dimensional MPEG-7 ColorLayout vector (see trace.moe-id), search by the vector directly. This is faster and saves bandwidth:

response = await api.search_vector("gwebWzth7oPe2UIubOJmozi1NDFp")  # base64 hash or 33 numbers

# Batch search: up to 10 vectors, each one costs 1 search credit
batch = await api.search_vectors([vector1, vector2])
for best in batch.best_results:
    print(best)

Account quota and usage

async def main():
    # Get your API key at https://trace.moe/account; without a key you are identified by your IP address
    async with TraceMoe(token="your_token_here") as api:
        me = await api.me()
        print(f"Priority: {me.priority}, concurrency: {me.concurrency}")
        print(f"Quota: {me.quota_used}/{me.quota} used in the last 24 hours, {me.quota_left} left")

        for slot in await api.usage("day"):
            print(slot.time, slot.total, slot.by_status)

Error handling and retries

Every API error has its own exception. All of them subclass TraceMoeAPIError:

from aiotracemoeapi import SearchQuotaDepleted, TraceMoe, TraceMoeAPIError

# Retry up to 3 times on concurrency limit, rate limit, full queue or overload errors
async with TraceMoe(max_retries=3, retry_delay=1.0) as api:
    try:
        response = await api.search("image.jpg")
    except SearchQuotaDepleted as e:
        print(f"Quota depleted: {e.quota_used}/{e.quota}")
    except TraceMoeAPIError as e:
        print(f"API error {e.status_code}: {e.text}")

Custom HTTP client or server

import httpx

async with httpx.AsyncClient(proxy="http://localhost:8080") as client:
    api = TraceMoe(client=client)  # your client is not closed by the wrapper
    ...

api = TraceMoe(base_url="https://trace.example.com")  # self-hosted trace.moe-api

Running Examples

uv run examples/console.py https://images.plurk.com/32B15UXxymfSMwKGTObY5e.jpg
uv run examples/console.py path/to/screenshot.jpg

Development

uv sync
make check  # ruff, mypy and pytest

See CHANGELOG.md for the list of changes, including how to migrate from 3.x.