kytmanov

kytmanov / synto

Public

More than just Karpathy’s LLM Wiki, 100% local with Ollama. Drop Markdown notes → AI extracts concepts → your Obsidian wiki auto-links and grows. Zero sharing. Your notes stay yours.

git-based-wiki karpathy knowledge-base llm-knowledge-base llm-pipeline
14
0
85% credibility
Found May 19, 2026 at 14 stars -- GitGems finds repos before they trend. Get early access to the next one.
Sign Up Free
AI Analysis
Python
AI Summary

Synto transforms your scattered notes into a self-improving, cross-linked wiki using local AI. You drop Markdown notes or PDFs into a folder, and it automatically extracts concepts and writes one article per topic. The wiki grows smarter with every note you add, supports asking questions in plain language, and can be exported as a portable knowledge pack for AI assistants. Everything runs locally by default, keeping your notes private.

How It Works

1
📚 You gather your notes

You collect all your research notes, book highlights, and ideas into one folder on your computer.

2
🤖 You connect your AI assistant

You set up a free local AI service so your notes never leave your machine — everything stays private.

3
✨ Your wiki writes itself

Synto reads through all your notes, finds the key concepts, and automatically creates beautifully linked articles about each topic.

4
👀 You review what was created

You flip through the generated articles, approve the ones you like, and give feedback on anything that needs more detail.

5
💬 You ask questions and get answers

Instead of scrolling through notes, you simply ask questions and your AI assistant answers from your own knowledge base.

6
How do you want to use it?
💻
Use it yourself

Your personal searchable knowledge base that works completely offline on your machine.

📦
Share with AI tools

Export everything as a portable pack that any AI assistant can read and reason about.

🎉 Your knowledge works for you

Every note you add makes your wiki smarter. Your ideas connect, compound, and stay organized forever.

Sign up to see the full architecture

5 more

Sign Up Free

Star Growth

See how this repo grew from 14 to 14 stars Sign Up Free
Repurpose This Repo

Repurpose is a Pro feature

Generate ready-to-use prompts for X threads, LinkedIn posts, blog posts, YouTube scripts, and more -- with full repo context baked in.

Unlock Repurpose
AI-Generated Review

What is synto?

Synto turns raw Markdown notes into a self-improving, cross-linked knowledge wiki using local LLMs. Drop notes into a folder, and the system analyzes them with a fast model, extracts concepts, and generates full articles using a heavier model—all on your own machine via Ollama. Each article links to related concepts via Obsidian-style `[[wikilinks]]`, and every note you add makes the wiki richer over time. The output is an agent-ready "pack" that Claude, Cursor, or any file-aware AI can read directly, with INDEX.json enabling fast concept lookup without embeddings or a vector database.

Why is it gaining traction?

The appeal is privacy-first AI that never sends your notes anywhere, combined with the intelligence of Karpathy's LLM Wiki concept without the sharing. The two-tier LLM architecture is clever: a small 4B model handles the fast, pattern-matching work of concept extraction, while a 14B+ model writes the actual articles—keeping the pipeline cheap and responsive on consumer hardware. Features like rejection feedback loops, incremental compilation, hand-edit protection, and a built-in MCP server make it genuinely useful for daily workflows, not just a demo. Supporting 20+ LLM providers (Ollama, LM Studio, vLLM, Groq, etc.) via OpenAI-compatible API means you're not locked into one setup.

Who should use this?

Developers and researchers who keep notes in Markdown and want to leverage AI to organize and link them without third-party services. Useful for learning new domains (e.g., building a wiki from course notes or technical books), maintaining architectural decision records that grow over time, or creating context layers for AI coding assistants. Researchers working with PDFs can import and segment documents with type-specific prompts. Anyone who wants Obsidian-like cross-linking but with AI-generated articles rather than manual creation.

Verdict

At 14 stars and v0.2.1, synto is early-stage but the codebase is thoughtful, the tests are thorough, and the feature set is already functional. The 0.8500000238418579% credibility score reflects solid engineering hygiene (type hints, pydantic models, comprehensive smoke tests, clear docs). Worth trying if you're comfortable with an experimental tool—if you already run Ollama locally, the setup is under five minutes and the workflow genuinely works.

Sign up to read the full AI review Sign Up Free

Similar repos coming soon.