Integration & Deployment

Four access methods — one of them fits you

From a one-line command to deep integration — choose freely by your team's stack

Command Line CLI

For: ops batch processing

Complete preprocessing, chunking and evaluation in a single command. Supports batch directories and resumable runs, and scripts into your existing data processing flow.

HTTP API

For: application backend integration

RESTful interfaces cover the full flow of preprocessing, knowledge base construction and evaluation, supporting sync/async, chunked uploads and SSE progress push.

C Shared Library SDK

For: deep C/C++/Java integration

Cross-platform .so/.dylib/.dll with C headers, precompiled for multiple platforms, embedded in your service process with zero network overhead.

Python / LangChain

For: AI application and data teams

Dependency-free Python SDK + LangChain connector: three lines of code load the enterprise knowledge base into a Document list, with metadata passed through.

How simple is it? Look at the code

From files to knowledge base — shorter than you think

Option 1: Command-line preprocessing (extract → clean → redact → compress, in one command)

# document preprocessing uglypearai preprocess --input contract.docx --output-dir out/ --clean --desensitize --compress # knowledge base construction (full-pipeline ingest, resumable) uglypearai chunk --mode ingest --input report.pdf --storage-dir ./rag_store --preset legal # run one evaluation pass uglypearai eval run --evalset my_set --report out/report.json

Option 2: Python + LangChain (most convenient for AI teams)

from uglypear_rag import UglyPearRAG from uglypear_rag.integrations.langchain_loader import UglyPearLoader client = UglyPearRAG("http://intranet-address:8080", api_key="sk-uglypear-***") loader = UglyPearLoader(client, user_tags=["dept:legal"]) docs = loader.load() # -> [Document], metadata such as chunk_id/version/permission labels passed through automatically

Option 3: HTTP API (callable from any language)

POST /v1/rag/preprocess # document → structured JSON + redacted copy POST /v1/rag/ingest # full-pipeline ingestion: chunking → enrichment → permissions → storage GET /v1/rag/documents # document and knowledge point queries, re-ingestion PATCH /v1/rag/.../validity # freshness governance: mark outdated/active POST /v1/rag/eval/run # trigger evaluation, produce quantified reports POST /v1/rag/feedback # negative-feedback attribution, badcase backflow

Deployment: from Docker to Indigenous-Stack Server Rooms

One command for standard environments, item-by-item adaptation for special ones

Standard Deployment (Docker)

# start the service (compression + full RAG pipeline) docker compose up -d # or run on bare metal RAG_STORAGE_DIR=/data/rag_store PORT=8080 ./uglypear-server

OCR/layout analysis models (ONNX) ship with the package and run locally; LLM summary enrichment connects to any OpenAI-compatible endpoint on the intranet (optional). Feature modules can be trimmed by feature flag: compression only, preprocessing only, or the full pipeline — take what you need.

Indigenous Technology Stack Environments

Supports Kylin/UOS operating systems, Phytion/Kunpeng CPUs (arm64 builds fully available) and Dameng/Kingbase databases, with fully offline installation. Verified in multiple projects compatible with China's indigenous technology stack. See details atChina's Indigenous Technology Stack Compatible Solution。

Connect to Your AI Stack

The platform outputs standard structured JSON and knowledge points with metadata, and can connect to:

Milvus Qdrant PGVector Elasticsearch LangChain LlamaIndex Dify Any in-house RAG

At the SDK level, FFI dynamic libraries are also provided (libcompressor.so + sdk.h) that C/C++/Java/Go can embed directly. See details atRust Compression SDKandDeveloper Center。

From Zero to Running: Five Steps

Usually measured in days, not months

Environment Preparation

Server specification list provided, delivered as Docker image or offline installer

Access Method Selection

Choose CLI / API / SDK per your team's stack — first test request within half a day

Pilot Documents Trial

Run a batch of real documents through the full pipeline and deliver parsing QC and cleaning reports

Strategy Tuning

Select chunking presets by document type and co-build industry evaluation sets with you

Production Go-Live

Connect vector stores and business systems, with dashboards monitoring continuously