Litho
Architecture docs drafted from the code itself: a Rust CLI whose LLM agents write a C4-style Markdown wiki with Mermaid diagrams, local Ollama included.
// LLM
16 entries tagged with "llm"
// COMPARISONS
1// ENTRIES
15 entries
anydoc
Word, Excel, PowerPoint and text PDFs to Markdown for an LLM, with no Python and no ML model: a Rust library, a Node CLI, a WebAssembly build. No local OCR.
Kwipu
Ask an Obsidian vault questions and get cited answers: a Python graph RAG on Ollama with an MCP server and a 3D view, though its default LLM runs in the cloud.
LLM Wiki
Compile documents once into a Markdown wiki an LLM keeps current, rather than re-reading them for every question. GPL-3.0 desktop app with graph and MCP server.
Unstract
Turn invoices, contracts and scans into JSON with an LLM, as an API or ETL pipeline. A self-hosted AGPL platform that hands OCR to a separate text extractor.
Hindsight
Long-term memory for AI agents: an LLM turns writes into facts and entities, four retrievers bring them back. MIT server on PostgreSQL, 1.0 since October 2026.
Markdownify MCP
Read PDFs, Office files, web pages and Git repos as Markdown from an MCP agent. MarkItDown and Python do the conversion, with no OCR, and audio goes to Google.
PaddleOCR
Pull text, tables and formulas out of PDFs and images as Markdown or JSON for an LLM: OCR and vision models under Apache 2.0, on PaddlePaddle by default.
Receipt OCR
Turns a photo of a receipt into JSON: merchant, date, line items, total. A Python library that hands the image to an OpenAI-compatible LLM. API key needed.
Meetily
Meeting notes transcribed on your own machine with Whisper or Parakeet. The text only reaches a remote LLM if you choose one for the summary. MIT desktop app.
Flowise
Drag-and-drop AI agents and RAG chains, each served over its own REST API. The project was archived in August 2026 and ships no releases or security fixes.
MarkItDown
Word, Excel, PowerPoint, PDF and EPUB files turned into Markdown an LLM can read, in Python. Images and scanned PDFs only become text by calling an LLM.
Agent Reach
Give an AI agent YouTube, Reddit, X and GitHub: a Python CLI picks a tool for each platform, installs it and checks it works. Most social sites need a login.
Dify
LLM chatbots, agents and RAG workflows built in a web console, each with its own API. Its modified Apache 2.0 license requires permission for multi-tenant use.
Hermes Agent
Hand tasks to an AI agent you can reach on Telegram, and it turns them into reusable skills. MIT code, any model, but shell commands run unsandboxed by default.