Engramly Parse · 面向大模型的文档引擎
Engramly Parse 是一款极速、无服务器化的解析引擎,把复杂 PDF — 表格、公式、多栏排版 — 转换为大模型可以直接消费的结构化 Markdown。
一次 API 调用 · 按页计费 · 每月免费 500 页
Revenue increased 23% year-over-year to $4.2B driven by strong demand in cloud services. Operating margins expanded to 34.1% reflecting scale efficiencies.
Note 14: The forward-looking statements herein are subject to risks...
## Financial Results FY2025 Revenue increased 23% year-over-year to $4.2B driven by strong demand in cloud services. Operating margins expanded to 34.1% reflecting scale efficiencies. | Line Item | Amount | |------------------|-----------| | Revenue | $4,200M | | COGS | $2,772M | | **Gross Profit** | $1,428M | | R&D | $588M | | **Op. Income** | **$840M** | > Note 14: Forward-looking statements > herein are subject to risks...
Real output — tables preserved as Markdown, headings structured, noise stripped.
Built for the LLM stack
The problem
Incumbent OCR services charge premium prices and return complex JSON you still need to post-process.
Most services process pages serially. A 100-page PDF means a 100× wait — unacceptable for real-time AI agents.
Legacy OCR returns bounding boxes, spans, and confidence scores. You still have to write a parser just to get usable text.
Your RAG pipeline deserves clean Markdown, not a weekend spent writing regex to reassemble shattered tables.
Tables
Financial reports, research papers, and legal documents are full of tables that break every OCR tool. Engramly Parse uses vision-based layout analysis to identify table structure and reconstruct perfect Markdown — including rowspan, colspan, and invisible borders.
| Region | Q1 | Q2 | Q3 | Q4 |
|---|---|---|---|---|
| N. America | 120 | 135 | 128 | 128 |
| Europe | 89 | 94 | — | 101 |
| APAC | 45 | 52 | 58 | 61 |
| Total | 254 | 281 | 186 | 290 |
Reading order
Academic papers, magazines, and government filings use multi-column layouts that naive OCR reads straight across. Engramly Parse detects column boundaries and reconstructs the natural reading flow — left column first, then right, with figures and captions placed correctly.
Formulas
Mathematical notation is the hardest thing to extract from PDFs. Engramly Parse decodes inline and display math into clean LaTeX — preserving subscripts, fractions, matrices, and Greek symbols exactly.
Three ways to parse
Whether you need a quick visual test, a Python integration, or a production HTTP endpoint — the same parsing engine powers all three.
Send a PDF file or URL. Get structured Markdown with tables, formulas, and headings. Scales from 1 page to 10,000 pages with zero configuration.
$ curl https://api.engramly.net/v1/parse \
-H "Authorization: Bearer $KEY" \
-F "file=@report.pdf" \
-F "output_format=markdown"
# → 47 pages parsed in 2.1s
# → clean Markdown with tablesThe official Python SDK wraps the API with type-safe methods, automatic retries, and streaming support. Drop it into your RAG pipeline in minutes.
from engram_pdf import Engram
client = Engram(api_key="zpka_...")
result = client.parse(
"financial_report.pdf",
mode="pipeline",
)
print(result.markdown)
# | Revenue | $4,200M |
# | COGS | $2,772M |No signup required. Upload any PDF and see the parsed Markdown output instantly. Compare the original document side-by-side with structured output.
Massively parallel inference
Every page of your document is processed in parallel across a serverless GPU fleet, then reassembled into a single Markdown output. No queues, no per-page wait. Scale to zero when idle. Pay only for compute used.
Built for developers
No CUDA setup, no GPU management, no post-processing scripts. Send a PDF, get Markdown that your LLM can read directly.
from engram_pdf import Engram
client = Engram(api_key="zpka_...")
# Parse a complex financial report
result = client.parse(
"10k_annual_report.pdf",
mode="pipeline",
)
# Feed directly into your RAG
for block in result.blocks:
vector_db.upsert(
text=block.content,
metadata={"type": block.type}
)from engram_pdf.integrations import (
EngramPDFLoader,
)
loader = EngramPDFLoader(
api_key="zpka_...",
mode="pipeline",
)
docs = loader.load("research_paper.pdf")
# Each doc has clean markdown
# with tables + formulas intact
chain.invoke({"docs": docs})$ curl https://api.engramly.net/v1/parse \
-H "Authorization: Bearer zpka_..." \
-F "file=@paper.pdf" \
-F "mode=pipeline" \
-F "output_format=markdown"
{
"pages": 47,
"elapsed": 2.1,
"markdown": "## Results\n\n| ...",
"blocks": [...]
}Pricing
One simple price. No request fees, no model surcharges, no integration tax. Scale from prototype to production without rewriting your billing forecast.
What 3,000 pages a month actually gets you
That's $0.003 per page at Pro — less than a third of what most legacy OCR APIs charge, and faster than anything that still streams pages one at a time.
14-day money-back guarantee on your first paid month · Refund Policy · Terms of Service
We'll send you an API key when your account is ready. No spam.