axoviq-ai/synthadoc
LLM ENGINE TURNING DOCUMENTS INTO STRUCTURED WIKI
Open-source engine that turns raw documents (PDFs, video, spreadsheets) into a structured local wiki on Markdown. Automatically builds cross-references, detects contradictions, and cites sources.
// KEY FEATURES
- Reads PDFs, video, spreadsheets, web pages, and AI session transcripts
- Automatically creates cross-references and detects contradictions
- Answers with source citations, stores everything in local Markdown
- Four interfaces: CLI, Obsidian, Web UI, and MCP for agents
// DEEP DIVE: HOW IT WORKS
Synthadoc ingests source files — PDFs, spreadsheets, presentations, web pages, images, video, Word docs, text, and JSONL transcripts — and runs them through an LLM to synthesize a unified wiki. Output is saved as Markdown files with automatic cross-references, contradiction flags, and orphan page markers; every answer cites its sources. Keyword indexing takes ~20 ms at 2,000 pages; semantic search puts the right page in the top 5 for 70% of queries; the knowledge graph visualizes thousands of pages in seconds.
Unlike classic RAG, Synthadoc needs no vector DB in production: all structure lives in readable files openable in Obsidian, Git, or any editor. The engine runs autonomously: an agentic maintenance cycle finds gaps, resolves conflicts, and updates articles without human intervention. Supports local models via Ollama and cloud (Gemini Flash), with cost audit trails, CI/CD hooks, and OpenTelemetry monitoring.
Requires Python and LLM access (API key or local server). Large wikis need resources for indexing and graph building; semantic search is enabled by default on new wikis. No built-in auth or RBAC — access is controlled at filesystem or reverse-proxy level. The MCP interface lets you plug the wiki as a tool into agents.
// USE CASES
Best for: Researchers, teams, and enterprises needing a local knowledge base without cloud
- Create a personal research wiki from PDFs, notes, and video in an hour
- Merge department knowledge into a single base with automatic conflict resolution
- Connect a local wiki as a tool to an AI agent via the MCP protocol
New repositories every 30 minutes
REDDYX AI scans GitHub 24/7 and ships the best AI/ML/Web3 projects to Telegram.
Join on Telegram