ITS-AI Python SDK

Typed, ergonomic Python client for the ITS-AI API with sync and async interfaces, robust error mapping, attempts, and helpful docs.

• Sync and async clients: ItsAIClient (requests) and AsyncItsAIClient (httpx)
• Strongly-typed results via dataclasses
• Attempts on 5xx and rate limits, with exponential backoff and Retry-After support
• Rich, typed error hierarchy mapped from API type and HTTP status
• Safe logging with masked API keys

Installation

pip install its-ai

Quick start

Single text analysis

from its_ai import ItsAIClient, AnalyzeTextResult

client = ItsAIClient(api_key="api_key")
try:
    result: AnalyzeTextResult = client.analyze_text("Your English text here.")
    print(result.answer)
finally:
    client.close()

With deep scan:

with ItsAIClient(api_key="api_key") as client:
    result = client.analyze_text("Your English text here.", deep_scan=True)
    print(result.answer, result.segmentation_tokens)

Batch analysis

from its_ai import ItsAIClient, AnalyzeBatchItemResult

texts = [
    "Short text",  # might trigger LowWords
    "A sufficiently long English text ...",
]

with ItsAIClient(api_key="api_key") as client:
    items: list[AnalyzeBatchItemResult] = client.analyze_batch(texts, deep_scan=False)
    for item in items:
        print(item.text, item.answer)

Chunking large batches automatically:

with ItsAIClient(api_key="api_key") as client:
    results = client.analyze_batch(texts, max_batch_size=50)

Plagiarism

A plagiarism check costs twice the words of an AI scan of the same text and can take minutes, so it uses its own generous timeout (PLAGIARISM_TIMEOUT) and is not retried — a retry cannot resume the abandoned scan, it starts a second one that is billed again. Pass timeout only to go higher.

with ItsAIClient(api_key="api_key") as client:
    result = client.check_plagiarism("Your text here.")
    print(result.score)                 # 0.0 original – 1.0 fully copied
    for source in result.results:
        print(source.score, source.link, source.title)
        for match in source.matches:    # the fragments that matched this source
            print(match.match_score, match.text_sentence, match.link)

Grammar & style

Unlike AI detection, the grammar endpoints are multilingual and accept short texts (from 20 characters). There is no score — the result is the list of issues.

with ItsAIClient(api_key="api_key") as client:
    result = client.check_grammar("She go to school every day.")
    print(result.language, result.stats.errors)
    for match in result.matches:
        print(match.severity, match.message, match.replacements)

    # Several texts at once — a failing text carries `error` instead of matches
    for item in client.check_grammar_batch(["First text ...", "Second text ..."]):
        print(item.error or item.stats.total)

report_id opens the web report, but the PDF certificate covers AI and plagiarism only — downloading it for a grammar-only scan returns 404.

PDF certificate

Every AI and plagiarism scan returns a report_id you can exchange for the PDF certificate. The endpoint takes no API key — the report_id is the secret — so the link is shareable.

with ItsAIClient(api_key="api_key") as client:
    result = client.analyze_text_v2("Your text here ...")
    pdf = client.download_report(result.report_id, lang="fr", tz="Europe/Paris")
    open("certificate.pdf", "wb").write(pdf)

lang and tz are optional (English / UTC by default). An unknown report_id, or one from a grammar-only check, raises NotFound.

Async usage

import asyncio
from its_ai import AsyncItsAIClient

async def main():
    async with AsyncItsAIClient(api_key="api_key") as client:
        res = await client.analyze_text("hello world", deep_scan=True)
        print(res)

asyncio.run(main())

Errors and attempts

The client raises typed exceptions derived from ItsAIError based on the API error type and HTTP status. Common ones include:

• ValidationError, AuthenticationFailed, PermissionDenied, NotFound, NotAcceptable
• Domain errors: LowWords, ManyWords, OnlyEnglish, RateLimitExceeded, etc.
• Transport failures: RequestTimeout, NetworkError

API errors arrive as <type>:<code> (e.g. validation:low_words, server:server) and are matched on both halves, so an unfamiliar code still lands on its category's class rather than on the bare ItsAIError.

Idempotent POSTs are retried up to 3 times on 5xx and on a rate limit, with exponential backoff and Retry-After respected. A rate limit arrives as HTTP 400 with code validation:rate_limit (not 429) and is raised as RateLimitExceeded; every other 4xx is final. Plagiarism checks are never retried — see below.

from its_ai import ItsAIClient, LowWords, ManyWords, OnlyEnglish, AuthenticationFailed

try:
    with ItsAIClient() as client:  # reads ITS_AI_API_KEY from env by default
        client.analyze_text("too short")
except LowWords as e:
    print("Text too short:", e.message)
except ManyWords:
    print("Text too long")
except OnlyEnglish:
    print("Only English is supported")
except AuthenticationFailed:
    print("Invalid/absent API key")

Configuration

• api_key: string, required (defaults from ITS_AI_API_KEY)
• base_url: defaults to https://api.its-ai.org (trailing slashes trimmed)
• timeout: default 10s (override per-call via timeout=)
• max_attempts: default 3 (5xx and rate limits)
• max_batch_size (batch-only): optional chunking of input texts

Headers are set automatically: User-Agent: its-ai-python-sdk/<version>, Accept: application/json, Content-Type: application/json.

Logging

The package uses Python's logging under the logger name its_ai. Enable DEBUG to see request URLs, status codes, and trimmed payloads. The api_key is masked.

import logging
logging.basicConfig(level=logging.DEBUG)

Environment

• ITS_AI_API_KEY – used by default if api_key is not passed.
• ITS_AI_E2E=1 – enable smoke tests to hit the real API in CI (optional).

Testing

Run unit tests:

python -m pytest -q

Run smoke (real API) tests when you have a valid key:

export ITS_AI_API_KEY="api_key"
export ITS_AI_E2E=1
python -m pytest -q

API Reference (brief)

Every method has an await-able twin with the same signature on AsyncItsAIClient.

AI detection

• analyze_text(text, deep_scan=False, *, timeout=None) -> AnalyzeTextResult — v1
• analyze_batch(texts, deep_scan=False, *, timeout=None, max_batch_size=None) -> list[AnalyzeBatchItemResult] — v1
• analyze_text_v2(text, *, timeout=None) -> AnalyzeTextV2Result — richer result (score, ai_percentage, probabilities, segments), always a deep scan
• analyze_batch_v2(texts, *, timeout=None, max_batch_size=None) -> list[AnalyzeBatchV2ItemResult] — per-text error instead of one error for the whole batch

Plagiarism

• check_plagiarism(text, *, timeout=None) -> PlagiarismResult

Grammar & style

• check_grammar(text, *, timeout=None) -> GrammarResult
• check_grammar_batch(texts, *, timeout=None, max_batch_size=None) -> list[GrammarBatchItemResult]

Reports

• download_report(report_id, *, lang=None, tz=None, timeout=None) -> bytes — the PDF certificate

Note that API access is an Enterprise-plan feature, enforced per request: a key issued on Enterprise stops working after a downgrade (PermissionDenied).

License

MIT

For API details and the hosted endpoint see https://api.its-ai.org.