PDF.chat API

Pregled

PDF.chat API je majhen vmesnik REST. Najprej vi POST dokument, ki ga zaužije in dobi delo z besedilom dokumenta in razčlenitev na stran (tekst, omejevalne polja, zaupanje). Potem pa ste POST vprašanja proti tej zaposlitvi in dobiti odgovore na podlagi dokumenta, vsaka navede stran, s katere je prišel. Dela s 5 stranmi ali manj vrniti v liniji; več delovnih mest se takoj vrne z pending status, ki ga pregledate do done.

• Osnovni URL: https://pdf.chat
• Dokumenti v: PDF, plus Word, PowerPoint, besedilo in slike (PNG, JPG, WEBP, GIF, BMP, TIFF)
• Chat odpovedanih: odgovore s straninimi citatami; transkripti skozi dogodek zgodovine
• Obdelano besedilo: txt, md, docx, pdf, csv, json
• Branje motorjev: cpu (hitro, tiskani docetiki) in vlm (premij AI, pisava, kompleksna postavitev, matematika)

Overitev

Preverite pri vas API token (Najdi jo na svojem stran računa) kot naslov Nosilca:

Authorization: Bearer YOUR_API_TOKEN

Lahko tudi predate ?api_token=… kot parametr za poizvedbo. Uporaba je merena v skladu s stransko saldo vašega računa.

Odpremljenost

POST /api/v1/ocr/, večdelni obrazec nalaganje.

curl -X POST https://pdf.chat/api/v1/ocr/ \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -F "file=@invoice.pdf" \
  -F "tier=vlm" \
  -F "language=auto"

Vrne nalogo. Za datoteke ≤5 strani je že to done z besedilom; večje datoteke se vračajo pending/processing, ocenite stanje opazovanega dogodka.

{
  "uuid": "9f2c1b7e4a...",
  "status": "done",
  "tier": "vlm",
  "language": "auto",
  "page_count": 1,
  "mean_confidence": 0.98,
  "text": "INVOICE\nAcme Corp\nTotal: 215.00 USD",
  "markdown": "# INVOICE\n\n**Acme Corp** ...",
  "pages": [ { "index": 0, "text": "...", "blocks": [ { "text": "...", "bbox": [x0,y0,x1,y1], "confidence": 0.98 } ] } ]
}

Dobi rezultat

GET /api/v1/ocr/<uuid>/, anketa do status je done ali failed.

curl https://pdf.chat/api/v1/ocr/9f2c1b7e4a.../ \
  -H "Authorization: Bearer YOUR_API_TOKEN"

Prenesi format

GET /api/v1/ocr/<uuid>/download/?format=md, izvaža rezultat. format je eden od txt, md, docx, pdf, csv, json.

curl -L "https://pdf.chat/api/v1/ocr/9f2c1b7e4a.../download/?format=docx" \
  -H "Authorization: Bearer YOUR_API_TOKEN" -o result.docx

Chat z dokumentom

Vprašaj o končanem delu. Odgovori so utemeljiti le v izvlečenem besedilu in navesti izvorno stran. Zahteva žeton računa, funkcija chat je račun-vrata.

POST /api/v1/chat/<uuid>/, TRUPLO JSON {"message": "your question"}.

curl -X POST https://pdf.chat/api/v1/chat/9f2c1b7e4a.../ \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"message": "What is the invoice total and due date?"}'

Vrača pomočnico sporočila z odgovorom in seznamom naštetih strani:

{"conversation": "a1b2…", "message": {
   "role": "assistant",
   "content": "The total is $42, due on March 3 (p. 1).",
   "citations": [{"page": 1, "cited_text": "The invoice total is $42…", "document_id": "9f2c1b7e4a…"}]
}}

GET /api/v1/chat/<uuid>/history/, Prinesi celoten zapis pogovora za službo.

Primeri kode

• Python
• Node.js
• cURL

import requests, time

BASE = "https://pdf.chat/api/v1"
H = {"Authorization": "Bearer YOUR_API_TOKEN"}

# 1. Upload a PDF
with open("contract.pdf", "rb") as f:
    job = requests.post(BASE + "/ocr/", headers=H, files={"file": f}).json()

# 2. Wait until it's ready to chat
while job["status"] in ("pending", "processing"):
    time.sleep(2)
    job = requests.get(f"{BASE}/ocr/{job['uuid']}/", headers=H).json()

# 3. Ask questions — every answer is cited to the page
ans = requests.post(f"{BASE}/chat/{job['uuid']}/", headers=H,
    json={"message": "What is the termination notice period?"}).json()
print(ans["message"]["content"])
print(ans["message"]["citations"])

import fs from "fs";

const BASE = "https://pdf.chat/api/v1";
const H = { Authorization: "Bearer YOUR_API_TOKEN" };

// 1. Upload a PDF
const form = new FormData();
form.append("file", new Blob([fs.readFileSync("contract.pdf")]), "contract.pdf");
let job = await (await fetch(`${BASE}/ocr/`, { method: "POST", headers: H, body: form })).json();

// 2. Wait until it's ready to chat
while (["pending", "processing"].includes(job.status)) {
  await new Promise(r => setTimeout(r, 2000));
  job = await (await fetch(`${BASE}/ocr/${job.uuid}/`, { headers: H })).json();
}

// 3. Ask questions — every answer is cited to the page
const ans = await (await fetch(`${BASE}/chat/${job.uuid}/`, {
  method: "POST", headers: { ...H, "Content-Type": "application/json" },
  body: JSON.stringify({ message: "What is the termination notice period?" })
})).json();
console.log(ans.message.content, ans.message.citations);

# 1. Upload a PDF
curl -X POST https://pdf.chat/api/v1/ocr/ \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -F "file=@contract.pdf"

# 2. Ask questions (use the uuid from step 1) — answers cited to the page
curl -X POST https://pdf.chat/api/v1/chat/UUID/ \
  -H "Authorization: Bearer YOUR_API_TOKEN" \
  -H "Content-Type: application/json" \
  -d '{"message": "What is the termination notice period?"}'

Parametri

& Omejitve napak

Vsaka stran obdelanih stroškov kreditov (1/stran na hitri ravni, več na premiji). Plačani načrti dvignejo zgornje meje na stran datoteke in dodajo prednost. Glej cene.

Pogosta vprašanja