TextSight API Examples

Runnable examples for the TextSight v2 API — AI content detection and rewriting — in Python, JavaScript, PHP and curl.

Every example here has been executed against the live API. No pseudocode, no dependencies beyond each language's standard library, and no framework to learn — each client is a single file you can copy into your own project.

git clone https://github.com/textsight/textsight-api-examples
cd textsight-api-examples
export TEXTSIGHT_API_KEY=sk_live_...      # from https://app.textsight.ai/api-keys

python3 python/detect.py                  # Python 3
node javascript/detect.mjs                # Node 18+
php php/detect.php                        # PHP 8+
./curl/examples.sh detect                 # curl

All four print the same thing. Official reference: textsight.ai/api-docs.html.

---

Read this before you parse a response

ai_probability is a fraction. The other scores are percentages.

Field	Range
ai_probability	0.0 – 1.0
humanization_score	0 – 100
confidence	0 – 100
sentences[].score	0.0 – 1.0

0.75 means 75% likely AI. Showing it to a user as "0.75% AI" — or multiplying the wrong field by 100 — is the single easiest mistake to make against this API. Every example here converts explicitly and prints the raw value alongside.

Get a key

Create one at app.textsight.ai/api-keys. Keys are shown once, at creation. Both auth forms work:

Authorization: Bearer sk_live_...      # documented form, used by these examples
x-api-key: sk_live_...                 # also accepted

⚠️ TextSight keys start with sk_live_, the same prefix Stripe uses for secret keys. Label them carefully if you handle both — they are indistinguishable at a glance. GitHub's secret scanner reads this shape as a Stripe key and will block a push containing one, including a fake one in a test fixture.

Keep keys server-side. Never ship one in client-side JavaScript.

Endpoints

Base URL: https://api.textsight.ai

Method	Path	What it does
POST	/v2/detect	Full analysis: score, verdict, reasoning, per-sentence breakdown
POST	/v2/score	The same scores without sentences or reasoning — smaller response
POST	/v2/rewrite	Rewrite the text, returning it plus the score of the rewrite

Bulk scanning (/v2/bulk) and job polling (/v2/jobs/<id>) are documented as planned and currently return 404. Do not build against them yet.

POST /v2/detect

{ "text": "..." }

Response:

{
  "humanization_score": 25,
  "ai_probability": 0.75,
  "verdict": "ai",
  "confidence": 85,
  "reasoning": "Analysis complete.",
  "sentences": [
    { "text": "...", "score": 0.75, "label": "ai" }
  ],
  "model": "gpu",
  "request_id": "req_5d73f8b7-f57a-464d-b690-8022a47558ea"
}

• verdict is human, mixed or ai.
• mixed is an inconclusive result, not a weak accusation. Short or ambiguous input lands here with low confidence. Treat it as "the detector could not commit" and present it that way to users.
• request_id identifies the call. Quote it in any support request.
• There is no minimum text length, but very short input produces a low-confidence mixed — as it should.

POST /v2/score

Same request. Returns humanization_score, ai_probability, confidence and request_id — no sentences, no reasoning. Use it when you only need the number.

POST /v2/rewrite

{ "text": "..." }

Response:

{
  "rewritten": "...",
  "humanization_score": 12,
  "ai_probability": 0.88,
  "score_reliable": true,
  "request_id": "req_2d627b17-5524-4ab7-9363-e82a443a546d"
}

Two things to get right:

1. The scores describe the rewrite, not your input. There is no before-and-after in one call. To show a genuine comparison, call /v2/detect on the input first — which is exactly what rewrite.py, rewrite.mjs and rewrite.php do here.
2. score_reliable: false means do not present those scores as fact. Render them as unavailable rather than showing a number the API has told you not to trust.

Errors

{ "error": { "code": "unauthorized", "message": "Invalid API key." } }

The error object is nested — error.code and error.message, not a top-level error string.

Status	code	Meaning
400	invalid_request	text missing or not a non-empty string
401	unauthorized	Key missing or rejected
429	—	Rate limited

Rate limits and quotas follow your plan tier; see textsight.ai/pricing.

What is in here

python/       textsight.py  + detect.py, score.py, rewrite.py     stdlib only
javascript/   textsight.mjs + detect.mjs, rewrite.mjs             Node 18+, built-in fetch
php/          textsight.php + detect.php, rewrite.php             ext-curl, ext-json
curl/         examples.sh                                          all three endpoints as raw curl

Each language has a small client file and one script per endpoint. CI checks that all three compile and that they fail identically when the key is missing or rejected.

A note on the rewrite endpoint

It rewrites text for readability and reports the detector's score honestly.

It is not an "undetectable" tool. No rewriter can guarantee a given detector's verdict — detectors change without notice, and none of these tools have access to the target model. Evading detection in an academic submission is misconduct at most institutions regardless of which tool produced the text.

If you want to know how reliable AI detection actually is, we publish our own detector's false-positive rate — including the results that do not flatter us — at textsight/ai-detector-benchmarks.

Licence

MIT. Copy these into your own project, commercial or otherwise.

Bug reports and examples in other languages are welcome. If a snippet here does not run, that is a bug — please open an issue.