Pdf Processing

PDF Processing

Quick start

Extract text:

from llama_cloud_services import LlamaParse

parser = LlamaParse(
    parse_mode="parse_page_with_agent",
    model="openai-gpt-4-1-mini",
    high_res_ocr=True,
    adaptive_long_table=True,
    outlined_table_extraction=True,
    output_tables_as_HTML=True,
    result_type="markdown",
    project_id=project_id,
    organization_id=organization_id,
)

result = parser.parse("./my_file.pdf")
documents = result.get_markdown_documents(split_by_page=True)

full_text = ""

for document in documents:
    full_text += document.text + "\n\n---\n\n"

For more detailed code implementations, see REFERENCE.md.

Requirements

The llama_cloud_services package must be installed in your environment:

pip install llama_cloud_services

And the LLAMA_CLOUD_API_KEY must be available as an environment variable:

export LLAMA_CLOUD_API_KEY="..."

LlamaParse

LlamaParse is a service for parsing and extracting text, charts, tables and images from unstructured files.

You can get started by creating simple scripts:

from llama_cloud_services import LlamaParse

parser = LlamaParse(
    parse_mode="parse_page_with_agent",
    model="openai-gpt-4-1-mini",
    high_res_ocr=True,
    adaptive_long_table=True,
    outlined_table_extraction=True,
    output_tables_as_HTML=True,
    result_type="markdown",
    project_id=project_id,
    organization_id=organization_id,
)

# sync
result = parser.parse("./my_file.pdf")

# sync batch
results = parser.parse(["./my_file1.pdf", "./my_file2.pdf"])

# async
result = await parser.aparse("./my_file.pdf")

# async batch
results = await parser.aparse(["./my_file1.pdf", "./my_file2.pdf"])

The result object is a fully typed JobResult object, and you can interact with it to parse and transform various parts of the result:

# get the llama-index markdown documents
markdown_documents = result.get_markdown_documents(split_by_page=True)

# get the llama-index text documents
text_documents = result.get_text_documents(split_by_page=False)

# get the image documents
image_documents = result.get_image_documents(
    include_screenshot_images=True,
    include_object_images=False,
    # Optional: download the images to a directory
    # (default is to return the image bytes in ImageDocument objects)
    image_download_dir="./images",
)

# access the raw job result
# Items will vary based on the parser configuration
for page in result.pages:
    print(page.text)
    print(page.md)
    print(page.images)
    print(page.layout)
    print(page.structuredData)

# you can also extract tables from the parse results
result = parser.get_json_result("./my_file.pdf")
tables = parser.get_tables(result)

See more details about the result object in the example notebook.

Using different parse pre-sets

In the following examples, you will find some sets of arguments that you can pass to LlamaParse. Please, make sure to use only these pre-defined sets of arguments when initializing LlamaParse.

from llama_cloud_services import LlamaParse

# Cost-Effective Mode
llama_parser = LlamaParse(
    # See how to get your API key at https://docs.cloud.llamaindex.ai/api_key
    parse_mode="parse_page_with_llm",
    high_res_ocr=True,
    adaptive_long_table=True,
    outlined_table_extraction=True,
    output_tables_as_HTML=True,
    result_type="markdown",
    project_id=project_id,
    organization_id=organization_id,
)

# Agentic Mode (Default)
llama_parser = LlamaParse(
    # See how to get your API key at https://docs.cloud.llamaindex.ai/api_key
    parse_mode="parse_page_with_agent",
    model="openai-gpt-4-1-mini",
    high_res_ocr=True,
    adaptive_long_table=True,
    outlined_table_extraction=True,
    output_tables_as_HTML=True,
    result_type="markdown",
    project_id=project_id,
    organization_id=organization_id,
)

# Agentic Plus Mode
llama_parser = LlamaParse(
    # See how to get your API key at https://docs.cloud.llamaindex.ai/api_key
    parse_mode="parse_page_with_agent",
    model="anthropic-sonnet-4.0",
    high_res_ocr=True,
    adaptive_long_table=True,
    outlined_table_extraction=True,
    output_tables_as_HTML=True,
    result_type="markdown",
    project_id=project_id,
    organization_id=organization_id,
)

Using with file object / bytes

You can parse a file object directly:

from llama_cloud_services import LlamaParse

parser = LlamaParse(
    parse_mode="parse_page_with_agent",
    model="openai-gpt-4-1-mini",
    high_res_ocr=True,
    adaptive_long_table=True,
    outlined_table_extraction=True,
    output_tables_as_HTML=True,
    result_type="markdown",
    project_id=project_id,
    organization_id=organization_id,
)

file_name = "my_file1.pdf"
extra_info = {"file_name": file_name}

with open(f"./{file_name}", "rb") as f:
    # must provide extra_info with file_name key with passing file object
    result = parser.parse(f, extra_info=extra_info)

# you can also pass file bytes directly
with open(f"./{file_name}", "rb") as f:
    file_bytes = f.read()
    # must provide extra_info with file_name key with passing file bytes
    result = parser.parse(file_bytes, extra_info=extra_info)

Using with SimpleDirectoryReader

You can also integrate the parser as the default PDF loader in SimpleDirectoryReader:

from llama_cloud_services import LlamaParse
from llama_index.core import SimpleDirectoryReader

parser = LlamaParse(
    parse_mode="parse_page_with_agent",
    model="openai-gpt-4-1-mini",
    high_res_ocr=True,
    adaptive_long_table=True,
    outlined_table_extraction=True,
    output_tables_as_HTML=True,
    result_type="markdown",
    project_id=project_id,
    organization_id=organization_id,
)

file_extractor = {".pdf": parser}
documents = SimpleDirectoryReader(
    "./data", file_extractor=file_extractor
).load_data()