axoviq-ai/synthadoc

Synthadoc: An open-source LLM knowledge compilation engine that turns raw documents into structured, local-first wikis. A transparent, human-readable alternative to traditional RAG, which can be self-managed and self-improved without the use of any tools.

What it solves

Synthadoc addresses the limitations of traditional Retrieval-Augmented Generation (RAG) systems, which typically summarize document chunks at query time. Instead, it compiles knowledge at the time of ingestion, synthesizing raw source documents into a structured, persistent wiki. This approach prevents contradictions from being blended silently, ensures claims are traceable to specific sources, and creates a permanent knowledge artifact that exists independently of the tool.

How it works

Synthadoc reads various source formats (PDFs, spreadsheets, PPTs, web pages, images, videos, Word files, TXTs, and AI session transcripts) and uses an LLM to synthesize them into a local Markdown-based wiki. It automatically builds cross-references using [[wikilinks]] and employs a 5-state lifecycle machine (draft $\rightarrow$ active $\rightarrow$ contradicted/stale $\rightarrow$ archived) to manage page status. The system includes an adversarial second-LLM pass to flag overstated claims and a linting process to detect orphan pages and broken citations.

Who it’s for

  • Solo researchers and indie hackers who need a personal research wiki using free or local LLM models.
  • Small teams requiring a centralized internal knowledge base that automatically resolves contradictions and scales with the organization.
  • Enterprises needing compliance-sensitive, local-first knowledge bases with audit trails and CI/CD integration via hooks.

Highlights

  • Ingest-time synthesis: Compiles sources into a wiki at ingestion rather than re-summarizing at query time.
  • Contradiction detection: Flags conflicting claims as status: contradicted for human review or auto-resolution.
  • Claim-level provenance: Provides inline citations (^[file:L-L]) linking every claim back to its source lines.
  • Local-first architecture: Stores data as plain Markdown files, ensuring no vendor lock-in and compatibility with tools like Obsidian.
  • Knowledge graph visualization: Generates a weighted knowledge graph with Louvain clustering to visualize relationships between pages.
  • Adversarial review: Uses a second LLM pass to flag unsupported superlatives and contestable facts.

Related

  • Project
    shannhk/llm-wikid

    An AI-maintained knowledge base for Obsidian that compiles raw sources into a structured wiki, avoiding RAG by pre-building cross-references and compounding knowledge over time.

  • Project
    atomicstrata/llm-wiki-compiler

    A knowledge compiler that turns raw sources into a structured, interlinked markdown wiki with citations and provenance for humans and AI agents.

  • Project
    SamurAIGPT/llm-wiki-agent

    LLM Wiki Agent is a local, agent‑driven system that ingests PDFs, DOCX, markdown, etc., extracts entities and concepts, builds an ever‑growing inter‑linked markdown wiki, flags contradictions, and generates a visual knowledge graph—all without a server or database.

  • Project
    nashsu/llm_wiki

    A desktop application that automatically transforms documents into a structured, interlinked personal wiki using LLMs, featuring a knowledge graph and automated deep research.

  • Project
    zosmaai/pi-llm-wiki

    A self-maintaining, Obsidian-compatible knowledge base for AI agents that transforms raw sources into a durable, interlinked wiki using the LLM Wiki pattern.