Datasheet Intelligence

Purpose

Prerequisites

uv

uv --version

uv

# macOS (Homebrew)
brew install uv

# Linux (official installer)
curl -LsSf https://astral.sh/uv/install.sh | sh

# Windows (WinGet)
winget install --id=astral-sh.uv -e

uv --version

Workflow

1. Place source datasheets in a folder (for example

   docs/datasheets/

   ).
2. Run

   scripts/ingest_docs.py

   with

   uv run --with docling python3

   .
3. Read outputs from

   .context/knowledge/

   :

   • <doc>/<doc>.md

     : normalized markdown

   • <doc>/<doc>.sections.jsonl

     : section chunks for retrieval

   • <doc>/<doc>.tables.md

     : table-focused markdown

   • <doc>/<doc>.meta.json

     : conversion metadata and validation info

   • <doc>/<doc>.docling.json

     : raw Docling structured export

   • <doc>/_images/*

     : extracted images

   • knowledge.index.json

     : corpus manifest
4. Use search/read helpers:

   • scripts/search_docs.py

     for corpus search

   • scripts/read_docs.py

     for focused section reads

Commands

# Ingest all supported datasheets from a directory
uv run --with docling python3 scripts/ingest_docs.py docs/datasheets --output-dir .context/knowledge

# Ingest only top-level files and skip OCR for faster runs
uv run --with docling python3 scripts/ingest_docs.py docs/datasheets --non-recursive --no-ocr

# Search and focused read
uv run python3 scripts/search_docs.py "SPI0 address" --knowledge-dir .context/knowledge
uv run python3 scripts/read_docs.py exynos_spi_v1 --anchor section-4-2

Operational Rules

• Use command-first guidance (above) instead of low-level converter internals.
• Prefer

  uv run

  commands and do not assume a global

  python

  alias.
• Keep output location at

  .context/knowledge

  unless user asks otherwise.
• Before the first run, ensure generated artifacts are ignored by git (

  .context/

  or

  .context/knowledge/

  in

  .gitignore

  ).
• Preserve image presence in markdown with

  ![image_ref](path)

  entries.
• Keep section anchors and normalize textual references like

  See Table 4.2

  into markdown links when anchor targets exist.
• Run on all provided files and summarize failures without stopping the whole batch unless explicitly requested.
• For detailed CLI options and execution presets, read

  references/execution-options.md

  .
• If

  uv

  is unavailable, follow the fallback instructions in

  references/execution-options.md

  .

Resources

• scripts/ingest_docs.py

  : main ingestion pipeline

• scripts/search_docs.py

  : chunk-level search helper

• scripts/read_docs.py

  : focused reader helper

• references/output-contract.md

  : output schema and retrieval contract

• references/execution-options.md

  : detailed runtime flags and command presets