Docling Reader

Pass DoclingReader to Knowledge.insert() to parse supported formats with Docling.

docling_reader.py

from agno.agent import Agent
from agno.knowledge.knowledge import Knowledge
from agno.knowledge.reader.docling_reader import DoclingReader
from agno.models.openai import OpenAIResponses
from agno.vectordb.pgvector import PgVector

db_url = "postgresql+psycopg://ai:ai@localhost:5532/ai"

knowledge = Knowledge(
    vector_db=PgVector(
        table_name="docling_documents",
        db_url=db_url,
    ),
)

agent = Agent(
    model=OpenAIResponses(id="gpt-5.2"),
    knowledge=knowledge,
    search_knowledge=True,
)

if __name__ == "__main__":
    knowledge.insert(
        name="Thai Recipes",
        url="https://agno-public.s3.amazonaws.com/recipes/ThaiRecipes.pdf",
        reader=DoclingReader(output_format="markdown"),
    )
    agent.print_response(
        "How do I make chicken and galangal in coconut milk soup?",
        markdown=True,
    )

Supported Input Groups

Group	Examples
Documents	PDF, DOCX, PPTX, Markdown, HTML, AsciiDoc, and LaTeX
Data	CSV, XLSX, XML, and Docling JSON
Images	PNG, JPEG, TIFF, BMP, and WebP
Audio	WAV, MP3, M4A, AAC, OGG, and FLAC
Video	MP4, AVI, and MOV

Run the Agent

Set up your virtual environment

uv venv --python 3.12
source .venv/bin/activate

Install dependencies

uv pip install -U "agno[docling]" openai pgvector psycopg sqlalchemy

Audio and video conversion also require:

uv pip install -U "docling[asr]" "docling-slim[format-video]"

These extras enable the audio and video capability groups described in Docling media processing. Install ffmpeg with your operating system's package manager. See FFmpeg downloads for platform packages.

Export the API key

export OPENAI_API_KEY=your_openai_api_key_here

Run PgVector

docker run -d \
  -e POSTGRES_DB=ai \
  -e POSTGRES_USER=ai \
  -e POSTGRES_PASSWORD=ai \
  -e PGDATA=/var/lib/postgresql \
  -v pgvolume:/var/lib/postgresql \
  -p 5532:5432 \
  --name pgvector \
  agnohq/pgvector:18

Run the agent

python docling_reader.py

Reader Parameters

Parameter	Type	Default	Description
`output_format`	`str`	`"markdown"`	Export format (`"markdown"`, `"text"`, `"json"`, `"yaml"`, `"html"`, `"html_split_page"`, `"doctags"`, `"vtt"`)
`converter`	`Optional[DocumentConverter]`	`None`	Custom Docling converter configuration
`format_options`	`Optional[dict]`	`None`	Format options dictionary for DocumentConverter
`chunking_strategy`	`Optional[ChunkingStrategy]`	`DocumentChunking()`	Strategy for chunking the document
`allowed_hosts`	`Optional[List[str]]`	`None`	Hostnames the reader is allowed to fetch from. See Restricting URL Fetches.

In the current implementation, Knowledge.insert(url=..., reader=DoclingReader(allowed_hosts=...)) downloads extension-bearing URLs before calling the reader. Validate the URL before ingestion because the reader's allowlist does not guard this path.

DoclingReader.async_read() runs the synchronous conversion in a worker thread.

Next Steps

Task	Guide
Compare available readers	Readers Overview
Configure chunking	Chunking
Parse PDFs with pypdf	PDF Reader