| name | deepresearch-pageindex |
| description | Use when managing a local DeepResearch knowledge base backed by PageIndex trees, including initializing a KB, adding named PDF/URL/Markdown/text/Word documents, listing documents, viewing a document tree, or reading one or more document sections by short section id, node, or range. |
DeepResearch PageIndex
Use the bundled CLI from this skill's scripts/ directory. It includes
deepresearch_kb.py, the PageIndex runtime scripts, and the local pageindex/
package copy:
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py --kb <kb_dir> <command> ...
If running from an installed skill outside the repository, use that installed
skill path instead. If Python dependencies are missing, install them from
scripts/requirements.txt.
Commands
Initialize a knowledge base:
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py --kb ./kb init
Initialize a named knowledge base under a parent directory:
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py --kb ./knowledge_bases init --name papers
This creates ./knowledge_bases/papers; use --kb ./knowledge_bases/papers for later commands.
Add a document. Always pass a stable --name; later commands use it instead of paths:
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py --kb ./kb add --name paper-a --source ./paper.pdf --model gpt-4o-2024-11-20
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py --kb ./kb add --name remote-paper --source https://example.com/paper.pdf --model gpt-4o-2024-11-20
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py --kb ./kb add --name notes --source ./notes.md
Supported inputs: PDF path, PDF URL, .md, .markdown, .txt, .docx, and .doc when local Microsoft Word automation is available.
List documents:
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py --kb ./kb list
Show one document tree:
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py --kb ./kb tree --name paper-a
Read parts from a document. --section-id/--sid, --node, and --range can be repeated in the same command:
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py --kb ./kb read --name paper-a --section-id a1b2c3d4 --node 0003 --range 5-7 --max-chars 6000
For PDFs, ranges are pages. For Markdown/text/Word-derived documents, ranges are line numbers.
Tree output includes both the PageIndex node ID and a short stable section ID
as [node_id|section_id]. Prefer section IDs when an LLM needs to revisit exact
document sections across steps.
Search literature from arXiv and/or OpenAlex:
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py search-lit --query "retrieval augmented generation"
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py search-lit --source arxiv --query "multi turn rag" --limit 10
python skills/deepresearch-pageindex/scripts/deepresearch_kb.py search-lit --source openalex --query "long term memory agents" --from-date 2023-01-01
Search defaults: --source all, --limit 10, and --from-date 2020-01-01.
For arXiv, prefer 2-3 keywords; long --source arxiv queries are rejected with
a reminder, while --source all skips arXiv and still queries OpenAlex.
Notes
- PDF tree generation uses PageIndex. Detected paper PDFs try MinerU heading extraction first; detected Chinese patent PDFs use the patent section preprocessor.
- API keys are environment variables, not committed files:
CHATGPT_API_KEY, optional OPENAI_BASE_URL, optional MINERU_API_TOKEN, optional MINERU_API_BASE, optional MINERU_OUTPUT_DIR.
- If a document name already exists, pass
--force to replace it.
