Skip to content

v0.3.0 — first PyPI release

Latest
Latest

Choose a tag to compare

View all tags
@andyliszewski andyliszewski released this 17 Apr 18:38
· 10 commits to main since this release
v0.3.0
afa78dc

First public release of grounding-ai on PyPI.

Grounding is a local-first document corpus pipeline for grounded AI agents. It converts PDF, EPUB, DOCX, and Markdown documents into a structured, searchable corpus with per-agent embedding indexes. Everything runs locally — no cloud APIs required.

Install

pip install grounding-ai

Requires Python 3.13.

What's in 0.3.0

Document pipeline

  • PDF / EPUB / DOCX / Markdown input; deterministic Markdown output with per-chunk YAML front matter
  • Per-chunk BLAKE3 hashing for provenance; SHA-1/SHA-256/BLAKE3 at document level
  • LangChain text splitters (1200/150 defaults); atomic file writes

Retrieval

  • Dense via FAISS + all-MiniLM-L6-v2 (384-dim, L2)
  • BM25 sidecar written alongside FAISS on every build and --incremental run
  • Hybrid retrieval via Reciprocal Rank Fusion (opt-in)
  • Cross-encoder reranking via bge-reranker-base (opt-in)
  • Page and section citations in retrieval output (`[slug, p.247, §3.2 Heading]`)
  • Retrieval evaluation harness with CI gate

Search surfaces

  • CLI tool (`grounding`)
  • MCP server for editor / agent integration
  • Agentic tool-calling loop for local LLMs (Ollama / LM Studio / any OpenAI-compatible endpoint)

Ingestion

  • Agent-filtered search via per-agent YAML configs with collection scoping
  • Staging watcher for drop-folder ingestion with automatic per-agent embedding updates

Links

Assets 2
Loading