txtai-xberg

Feed Xberg document extraction into txtai. XbergPipeline is a plain callable that extracts text and metadata from 91+ formats — running OCR where needed — and flattens the result into documents ready for txtai.Embeddings.index. When you enable Xberg's native chunking, each chunk becomes one embedding-sized segment instead of a single blob.

Install

pip install txtai-xberg

Requires Python 3.10+. Install the txtai extra (pip install "txtai-xberg[txtai]") if txtai isn't already in your environment.

Extract

Call the pipeline with a path or a list of paths. A string returns one document; a list returns documents in input order. Batches run through Xberg's extract_batch in a single native call.

from txtai_xberg import XbergPipeline

pipeline = XbergPipeline()

doc = pipeline("report.pdf")
print(doc["content"])              # extracted markdown
print(doc["metadata"]["title"])    # source, mime_type, title, authors, languages, page_count

docs = pipeline(["report.pdf", "notes.docx"])

Index into txtai

Use to_documents to get (id, text, tags) tuples and hand them straight to Embeddings.index. Enable Xberg chunking so segments arrive sized for the model, with heading and page context in each document's tags.

from txtai import Embeddings
from txtai_xberg import XbergPipeline
from xberg import ChunkingConfig, ExtractionConfig

pipeline = XbergPipeline(
    config=ExtractionConfig(chunking=ChunkingConfig(max_characters=1000, overlap=200)),
)
documents = pipeline.to_documents(["report.pdf", "notes.docx"])

embeddings = Embeddings(path="sentence-transformers/all-MiniLM-L6-v2", content=True)
embeddings.index(documents)

for result in embeddings.search("quarterly revenue", 3):
    print(result["id"], result["text"])

Without a chunking config, to_documents emits one document per file. Chunk ids are "<source>#<chunk_index>".

Configure extraction

Pass any Xberg ExtractionConfig to control OCR, output format, chunking, and concurrency.

from xberg import ExtractionConfig, OcrConfig

pipeline = XbergPipeline(
    config=ExtractionConfig(
        output_format="markdown",
        ocr=OcrConfig(language="eng"),
        force_ocr=True,
        max_concurrent_extractions=8,
    ),
)

Async

Already inside an event loop? Use the async methods — await pipeline.acall(paths) and await pipeline.ato_documents(paths). The synchronous __call__ and to_documents wrap these with asyncio.run and must not run inside a running loop.

Errors

Per-input failures in a batch surface as an ExtractionFailedError, whose errors attribute holds Xberg's ExtractionErrorItem objects (each with an index, source, code, and message).

For the full API, see the Xberg documentation.

Metadata

Release files for txtai-xberg 1.0.0rc38

For a detailed explanation of source distributions (sdists) and built distributions (wheels), please see the package formats documentation.

Source distribution (sdist)

Source distribution for txtai-xberg 1.0.0rc38
File	Size	Uploaded
txtai_xberg-1.0.0rc38.tar.gz	9.1 kB	Jul 25, 2026	Details

Built distribution (wheel)

Table of built distributions (wheels) for txtai-xberg 1.0.0rc38
File	Interpreter	ABI	Platform	Reset
txtai_xberg-1.0.0rc38-py3-none-any.whl 7.0 kB Jul 25, 2026	Python 3	none	any	Details

Total release size: 16.1 kB

Release files / txtai_xberg-1.0.0rc38.tar.gz

Download URL	txtai_xberg-1.0.0rc38.tar.gz
Size	9.1 kB
Tags	Source
SHA-256 checksum How to use checksums	`31755fb5562a64af773f60315a575b797e8a91e9242481071f1a6a7c15fc0136`
BLAKE2b-256 checksum How to use checksums	`1892086378723f09849f7ff4cbd61a0e28294a78d96f3a59eaa2341ee4d95bea`
Upload date	Jul 25, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`uv/0.11.32 {"installer":{"name":"uv","version":"0.11.32","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}`

Release files / txtai_xberg-1.0.0rc38-py3-none-any.whl

Download URL	txtai_xberg-1.0.0rc38-py3-none-any.whl
Size	7.0 kB
Tags	Python 3
SHA-256 checksum How to use checksums	`d3c38f4171f27143f2f7e60b9105406a353ed32c8e949b9f92d5ad72de958e24`
BLAKE2b-256 checksum How to use checksums	`2718b7553fe4bf7ef20b16fa5801f6f39fcc06b66bd65c092a4a004ee4016abe`
Upload date	Jul 25, 2026
Uploaded using Trusted Publishing? What is trusted publishing?	Yes
Uploaded via	`uv/0.11.32 {"installer":{"name":"uv","version":"0.11.32","subcommand":["publish"]},"python":null,"implementation":{"name":null,"version":null},"distro":{"name":"Ubuntu","version":"24.04","id":"noble","libc":null},"system":{"name":null,"release":null},"cpu":null,"openssl_version":null,"setuptools_version":null,"rustc_version":null,"ci":true}`