apunkt/PageIndex
1
0
Fork 0
mirror of https://github.com/VectifyAI/PageIndex.git synced 2026-06-21 20:18:09 +02:00
Code Issues Projects Releases Packages Wiki Activity

fix(filesystem): remove cat node reads

Browse source 
Return nested PageIndex structure JSON from cat --structure and keep content reads page-based only. Remove the cat --node command surface, related limits, prompts, and structure-text fallback.
This commit is contained in:
BukeLy 2026-06-01 00:05:19 +08:00
parent e368562e03
commit d0c0c67a39
6 changed files with 56 additions and 277 deletions
Download patch file Download diff file Expand all files Collapse all files

4
examples/pifs_demo.py
View file

@ -5,7 +5,7 @@ This mirrors examples/agentic_vectorless_rag_demo.py, but exposes a corpus
through the PageIndex FileSystem shell instead of direct PageIndex document
tools. The agent receives one read-only bash-like PIFS tool and must retrieve
evidence through commands such as ls, tree, browse, find, grep, cat <path>
--structure, cat <path> --page, and cat <path> --node.
--structure, and cat <path> --page.
The demo registers supported files under examples/documents. When a matching
examples/documents/results/*_structure.json file exists, it is loaded into the
@ -81,7 +81,7 @@ Retrieval strategy:
browse -R /documents "Federal Reserve supervision regulation"
- browse returns file candidates only; it is not folder semantic recall.
- After browse returns candidates, verify evidence with grep, cat <path>
--structure, cat <path> --node, or cat <path> --page before answering.
--structure, or cat <path> --page before answering.
- Use find --where only with JSON metadata DSL, for example:
find /documents --where '{"file_format":"pdf"}'
- Use grep -R only for lexical evidence; do not treat semantic candidates as