ugraph

What it is

ugraph is a local-first CLI for builders who want knowledge they can trust. Copy text or drop a YouTube playlist URL, run one command, and get a plain Markdown knowledge base with claims checked against source text — not vibes-only summaries.

Designed for Obsidian vaults and agent workflows. No opaque RAG product, no vendor lock-in.

Tagline: Local knowledge that agents can trust.

Why it exists

Most “AI knowledge” tools collapse into chat dumps: hard to audit, easy to forget, locked to someone else’s database. ugraph keeps the opposite contract — filesystem-native Markdown, content-addressed chunks, and citation-hard extraction before a claim becomes a concept.

What it does

• One capture habit — copy text or a YouTube playlist URL, run ugraph, confirm, done
• YouTube → structured knowledge — playlist ingest with transcripts, resume, and inspectable status
• Citation-hard extraction — quotes gated against source text before they land as candidates
• Filesystem-native KB — plain Markdown folders Obsidian and agents already understand
• Local-first — run on Ollama; sources and indexes stay on your machine
• Agent-ready graph — concepts and links that compound over time

Try it

uv tool install ugraph-kit

PyPI package: ugraph-kit · Source: github.com/saran-io/ugraph

Status

Alpha. Ingest and provenance are the core strength today. Ask / retrieval is next — intentionally not oversold.

Stack

• Python CLI (ugraph-kit)
• Local LLMs via Ollama
• Markdown-first knowledge layout (Obsidian-friendly)
• Provenance and citation checks on extracted claims