High-fidelity document parsing for AI, self-hosted
Structure, tables and layout preserved — not just text. No cloud account: docker run and go.
-
< 5min
from zero to production document parsing -
-40%*
fewer LLM tokens with DocLang
Structure, tables and layout preserved — not just text. No cloud account: docker run and go.
< 5min
from zero to production document parsing-40%*
fewer LLM tokens with DocLang From a raw document to clean DocLang or JSON, entirely inside your own infrastructure.
Pull the container and run it with your license file.
$ docker run -p 8080:8080 \
-e FINEPARSER_LICENSE_DATA="$(cat acme.fineparserlicense)" \
abbyyteam/fineparserPOST a document file to the API endpoint.
curl -X POST localhost:8080/parse \
-F file=@form.pdf \
-F outputType=doclangGet back DocLang or JSON
POST a file to /parse, get the parsed document back. Plain HTTP, so nothing to install and no SDK to keep up to date.
Writing the integration with Claude Code or Codex? Point it at the ABBYY docs MCP server and it writes against the real API. See how →
DocLang's markup maps cleanly to LLM tokens. Extend your tokenizer with it, and the same document costs up to 40% less to feed into your model.*
Extend any Hugging Face tokenizer with DocLang's vocabulary in a few lines:
from transformers import AutoTokenizer
import doclang.tokenization as dt
tokenizer = AutoTokenizer.from_pretrained("openai/gpt-oss-20b")
tokenizer.add_special_tokens(
{"additional_special_tokens": dt.get_special_tokens()}
)Real reductions across two production-grade tokenizers:
granite-4.0
gpt-oss-20b
*Full methodology and benchmark details in the FAQ.
FineParser is made by ABBYY, the document recognition company founded in 1989, named a Leader in the 2026 Gartner® Magic Quadrant™ for Intelligent Document Processing Solutions. Learn more about ABBYY →
All plans are self-service monthly subscriptions — except Enterprise. Pull the Docker image and parse your first document in under five minutes.
Processing stops when the monthly page limit is reached. No overages — upgrade or wait for the next billing period.
Evaluate FineParser on real documents, on your own infrastructure.
1,000 pages/month
For developers moving from evaluation to early production workloads.
25,000 pages/month
For teams running FineParser in production pipelines at meaningful scale.
250,000 pages/month
For organisations processing high document volumes with business support needs.
1,000,000 pages/month
Full FineReader Engine SDK. Air-gapped, unlimited volume, enterprise SLAs.
The FineReader Engine behind FineParser is trusted by Volkswagen and enterprises worldwide.
Every self-service tier is community-supported. Report a bug, read the source, or ask the team and other users directly — all in the open, on GitHub.
Hit a bug or a document that parses wrong? Open an issue with a sample and we'll take a look.
Open an issue →Questions about deployment, output formats or tuning for your documents — ask here.
Visit the repo →Examples and the DocLang tooling, including the doclang package.
Enterprise customers get ABBYY Business Support with defined SLAs. Compare support levels →
FineParser is a self-hosted, Docker-native document-parsing engine for AI applications, powered by ABBYY FineReader Engine.
Yes — FineParser runs entirely on your own infrastructure via docker run. Sign up to get your .fineparserlicense file, pass its contents to the container in the FINEPARSER_LICENSE_DATA environment variable, and you're parsing right away. No key to activate and nothing to mount.
FineParser starts free — 1,000 pages/month for a year. Paid plans run $119–$2,000/month for 25,000 to 1,000,000 pages, and Enterprise offers custom, volume-based licensing with the full FineReader Engine SDK. See pricing for the full tier and feature comparison.
Three: DocLang, JSON and TXT. DocLang is the structure-preserving default built for LLMs; JSON carries the same structure with pixel-accurate bounding boxes.
Yes. DocLang's markup is a controlled vocabulary built to map cleanly to LLM tokens, and extending your tokenizer with its special tokens cuts token consumption by roughly 40% on average.
*Methodology: benchmarked across 8,030 DocLang documents, comparing each tokenizer's base vocabulary against the same tokenizer extended with DocLang's special tokens via the open-source doclang package (doclang.tokenization.get_special_tokens()). Results: -41% mean / -40% median on granite-4.0, and -45% mean / -45% median on gpt-oss-20b. Actual savings depend on your documents and choice of base tokenizer. See DocLang speaks fewer tokens for the integration code.
200+ languages, including Latin, Cyrillic, CJK, and Arabic scripts.
FineParser only requires a connection to a single ABBYY server for subscription validation. The Enterprise tier supports full air-gapped deployment with zero connectivity.