Skip to content

InferX Knowledge Base Toolkit How to

Jump to bottom
Jason edited this page May 20, 2026 · 1 revision

Overview

This toolkit converts source material into Markdown that is easier to use for retrieval, summarization, and prompt construction.

It provides a containerized document conversion workflow using inferx/knowledgebase:v0.10.

Main Use Cases

Document Conversion

Use the container when you want to process files from /input and generate Markdown artifacts in /output.

Supported input types:

  • PDF
  • DOCX
  • PPTX
  • HTML / HTM
  • Markdown
  • Text

The container walks /input recursively and processes all supported files it finds.

Typical Workflows

1. Convert Local Documents

sudo docker run --rm \
  -v /home/brad/test/input:/input \
  -v /home/brad/test/output:/output \
  -e "API_KEY=YOUR_API_KEY_HERE" \
  inferx/knowledgebase:v0.10 \
  base_url=https://model.inferx.net/funccall/tn-a3t79iogb2/endpoints/Qwen3-Coder-Next-FP8/v1 \
  api_key=YOUR_API_KEY_HERE \
  model=Qwen/Qwen3-Coder-Next-FP8

Output Artifacts

Conversion Output

  • merged.md: original output with boundaries and summaries
  • optimized.md: compressed output for KV-cache-oriented usage
  • llm.md: prompt-ready version with instructions and citation guidance

LLM-Ready Format

The llm.md output is intended to be used directly in prompts. It includes:

  • a system instruction block
  • citation rules
  • section-preserving formatting
  • contextual handling for diagrams and partially rendered formulas

Expected citation form:

  • [bitcoin.pdf, Section 4 - Proof-of-Work]
  • [bitcoin.pdf, Section 11]
  • [bitcoin.pdf, Section 5, Step 3]

Avoid citations that omit the filename or section reference.

Configuration Reference

Container Arguments

  • base_url: LLM endpoint URL
  • api_key: API key for authentication
  • model: model identifier, defaulting to Qwen/Qwen3-Coder-Next-FP8

Environment Variables

  • API_KEY: alternative way to provide the API key

Performance Notes

  • Conversion is roughly 10 seconds per file
  • Lossless compression is effectively immediate
  • LLM-ready formatting is produced as part of normal output generation

Operational Notes

  • Lossless compression is deterministic and safe, but token reduction is small
  • OCR models are preloaded into the Docker image

When To Use What

Use this image for local document sets that need Markdown conversion and prompt-ready output.

Clone this wiki locally