Quickstart

Create a key, score a paragraph, draft a passage and check it. About five minutes.

1Create a key

Listing 1. shell

export ETAL_API_KEY="etal_..."

New accounts have a monthly allowance and a rate limit (see Models and limits). Nothing is billed.

2Score a paragraph with Sic

Listing 2. shell

curl https://api.etal-labs.com/v1/detect \
  -H "Authorization: Bearer $ETAL_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "sic", "text": "Notably, these findings underscore the need for larger, well-powered trials. It is important to note that adherence declined over time."}'

The reply has one entry per sentence, the two thresholds, and the measured error rates:

Listing 3. json

{
  "id": "dt_…", "object": "detection", "model": "sic-2026a",
  "document": {"max_score": 0.9682, "above_strict": false, "above_sensitive": true,
               "sentences": 2, "sentences_above_strict": 0, "sentences_above_sensitive": 2},
  "thresholds": {"strict": 0.9946, "sensitive": 0.9544},
  "sentences": [
    {"start": 0, "end": 76, "text": "Notably, these findings …", "score": 0.9682, "label": "above_sensitive"},
    {"start": 77, "end": 135, "text": "It is important to note …", "score": 0.9676, "label": "above_sensitive"}
  ],
  "error_rates": {"human_texts_flagged": {"strict": {"research_papers": {"flagged": 4, "n": 7950, "rate": 0.0005}, "…": "…"}}, "…": "…"},
  "usage": {"words": 20, "cost_usd": 0.0002}
}

A score above a threshold is evidence. It is not a finding about who wrote the text; read Responsible use before you act on one.

3Draft a passage with QED

Give QED the points to cover, one per line, and two or three passages you have written so that it can follow your voice. Put every fact, number and citation you want in the notes. QED adds none of its own.

Listing 4. python

import os, httpx

r = httpx.post(
    "https://api.etal-labs.com/v1/write",
    headers={"Authorization": f"Bearer {os.environ['ETAL_API_KEY']}"},
    timeout=120,
    json={
        "model": "qed",
        "task": "compose",
        "title": "Sleep deprivation and memory consolidation",
        "brief": "- 64 adolescents and 60 adults, crossover design\n"
                 "- next-morning recall fell by 21% vs 12% (interaction p = 0.008)\n"
                 "- the age difference was gone at 48 h (p = 0.21)",
        "voice": {"exemplars": [open("my_paragraph_1.txt").read(),
                                open("my_paragraph_2.txt").read()]},
        "target_words": 120,
    },
)
passage = r.json()["text"]

4Check it against your notes

QED can still drop a number or sharpen a hedge. compare lists what differs between the source and the output. The deterministic checks are free; "judge": true adds a claim-by-claim reading for about half a cent.

Listing 5. python

check = httpx.post(
    "https://api.etal-labs.com/v1/compare",
    headers={"Authorization": f"Bearer {os.environ['ETAL_API_KEY']}"},
    timeout=120,
    json={"source": brief, "output": passage, "kind": "draft", "judge": True},
).json()

if not check["pass"]:
    for w in check["warnings"]:
        print(w)          # e.g. {"type": "numbers_missing", "items": ["0.008"]}

5Say that you used it

Text written with QED is AI-assisted, and most journals ask you to say so. POST /v1/disclosure drafts the statement from your own recent calls, in the form a journal asks for. Edit it; it is a draft, not a certificate.