axoviq-ai/synthadoc

# Synthadoc

      .-+###############+-.
    .##                   ##.
   ##    .----.   .----.    ##
  ##    /######\ /######\    ##
  ##    |######| |######|    ##
  ##    | [SD] | | wiki |    ##
  ##    |######| |######|    ##
  ##    \######/ \######/    ##
   ##    '----'   '----'    ##
    '##                   ##'
      '-+###############+-'
S Y N T H A D O C
    Community Edition  v1.3.4
  ────────────────────────────────
  Domain-agnostic LLM wiki engine

Document version: v1.3.4

Engineered for solo users and enterprises alike, providing a domain-specific knowledge base that scales seamlessly while maintaining accuracy through autonomous self-optimization.

Built for individuals, small teams, and large organizations who need a knowledge base that stays accurate as documents accumulate.

Synthadoc reads your raw source documents — PDFs, spreadsheets, PPTs, web pages, images, videos, Word files, TXTs, and AI session transcripts (.jsonl) — and uses an LLM to synthesize them into a persistent, structured wiki. Cross-references are built automatically, contradictions are detected and surfaced, orphan pages are flagged, and every answer cites its sources. Outputs are stored as local Markdown files, ensuring seamless integration and autonomous management within Obsidian or any wiki-compliant ecosystem.

---

▶ From Documents to Wiki

▶ Four Interfaces: CLI, Obsidian, Web UI & MCP

▶ Agentic Maintenance Workflow

📝 Blogs & Media — YouTube · Coderlegion · DEV.to · Medium

📂 End-to-end Example — AquaFlow Capital M&A due diligence walkthrough

---

Table of Contents

• Who Is It For?
• Inspiration and Vision
• Problems Addressed
• Why Synthadoc?
• Architecture
• What's Included
• Installation
• Quick-Start Guide
• Creating Your Own Wiki
• Configuration
• Command Reference by Use Case
• Administrative Reference
• Understanding Logs and the Audit Trail
• Customization
• Links

Who Is It For?

Synthadoc scales from a single researcher to a company-wide knowledge platform:

| Team size | Typical use case | | ----------------------- | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Solo / 1–2 people | Personal research wiki, freelance knowledge base, indie hacker documentation - run it free on Gemini Flash or a local Ollama model with zero ongoing cost | | Small team (3–20) | Centralized internal knowledge base for startups and departments that aggregates diverse individual data sources into a unified, high-integrity wiki. The system automatically resolves contradictions and scales autonomously, ensuring organizational intelligence grows in tandem with your team | | Medium / enterprise | Compliance-sensitive knowledge bases that must stay local; per-department wikis on separate ports; audit trail for every ingest and cost event; hook system for CI/CD integration; OpenTelemetry for ops dashboards |

No cloud account. No vendor lock-in. The wiki is plain Markdown — open it in any editor, back it up with git, sync it with any cloud drive.

---

Inspiration and Vision

"The LLM should be able to maintain a wiki for you."

— Andrej Karpathy, LLM Wiki gist

Most knowledge-management tools retrieve and summarize at query time. Synthadoc inverts this: it compiles knowledge at ingest time. Every new source enriches and cross-links the entire corpus, not just appends a new chunk. The wiki is the artifact — readable, editable, and browsable without any tool running.

Long-term alignment:

| Direction | How Synthadoc moves there | | ------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | | Agent orchestration | Orchestrator dispatches parallel ingest, query, and lint sub-agents with cost guards and retry backoff | | Sub-agent skills/plugins | Featuring a 3-tier lazy-load capability system, the platform allows for the injection of custom skills and hooks via a plug-and-play interface, ensuring core stability is never compromised during extension | | LLM wiki vs. RAG | Pre-compiled structured knowledge beats query-time synthesis for contradiction detection, graph traversal, and offline access | | CLI / HTTP | A unified interface via CLI and RESTful endpoints, the system streamlines full-spectrum integration: from data ingestion and querying to automated linting, security auditing, and job orchestration | | Local-first | All data stays on your machine; localhost-only network binding; no cloud dependency except the LLM API itself | | Provider choice | LLM backends including free-tier Gemini and Groq, paid Anthropic/OpenAI/DeepSeek/MiniMax/Qwen (DashScope), local Ollama and Qwen, and coding-tool CLI providers (Claude Code, Opencode) — no API key required if you already have a subscription |

---

Problems Addressed

RAG retrieves document chunks at query time. Synthadoc compiles knowledge at ingest — synthesising sources into a linked, audited wiki graph so contradictions are caught, claims are traced to sources, and the artifact survives outside the tool.

| Problem | Synthadoc approach | | ----------------------------------------- | -------------------------------------------------------------------------------------------------------------------------------- | | Contradictions blended silently | Ingest-time conflict detection; page flaggedstatus: contradicted; auto-resolve or queue for human review | | No links between related content | [[wikilinks]] auto-built on every ingest pass; weighted graph (wikilink + co-source signals) with Louvain clustering in web UI | | Orphan pages never surfaced | Lint reports unreferenced pages with ready-to-paste index entries | | LLM output can be overconfident | Adversarial second-LLM pass flags overstated claims, unsupported superlatives, and contestable facts per page | | Claims lack source traceability | ^[file:L-L] citation on every claim; Source Viewer in Obsidian; PDF page resolution; broken-citation lint | | Knowledge lifecycle invisible | 5-state machine (draft → active → contradicted / stale → archived); auto-transitions via lint; immutable event log | | Repeat ingest is expensive | 3-layer cache (embedding, LLM, provider prompt) — repeat lint on unchanged pages costs near-zero tokens | | Knowledge locked in proprietary tools | Plain Markdown + YAML frontmatter; OKF v0.1 compatible; fully offline-readable in any editor | | Wiki structure drifts with growth | scaffold regenerates index, AGENTS.md, and purpose.md from current wiki state without touching linked pages | | Migration requires full re-ingestion | Single-zip backup + restore with port/domain rewriting; no re-ingestion needed | | Cost and compliance exposure | Localhost-only; per-job token+cost log; configurable soft-warn and hard-gate thresholds |

Citation quality: Generated pages include inline citations linking every claim to its source lines. Pages without citations trigger a model-compatibility warning — use Gemini 2.5 Flash or higher for reliable citation annotation.

> Active page protection: Pages promoted to active status are protected — sources that contradict them are flagged for review rather than overwriting the human-reviewed content.

---

Why Synthadoc?

Every Yes below is a built-in feature — no add-ons or upgrades required.

Knowledge Quality

| Capability | Synthadoc | Typical RAG | NotebookLM | Notion AI | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | --------- | ----------- | ---------- | --------- | | Ingest-time synthesis — sources compiled into the wiki at ingest; not re-summarised at query time | Yes | No | Partial | No | | Contradiction detection & resolution — conflicting claims flagged status: contradicted; auto-resolve available; full conflict history | Yes | No | No | No | | Adversarial claim review + gate — concurrent second-LLM pass flags overstated claims, unsupported superlatives, and contestable facts per page; configurable gate auto-demotes pages that exceed the warning threshold to contradicted | Yes | No | No | No | | Claim-level provenance — ^[file:L-L] citation on every claim; Source Viewer in Obsidian; PDF page resolution; broken-citation lint | Yes | No | Partial | No | | 5-state lifecycle machine — draft → active → contradicted / stale → archived; auto-transitions via lint; immutable event log; cascade link cleanup on archive; only active and stale pages enter the BM25 search index | Yes | No | No | No | | Pre-LLM source sanitizer — strips zero-width chars, bidi overrides, hidden HTML, and instruction-override phrases before any LLM call | Yes | No | No | No |

Knowledge Structure

| Capability | Synthadoc | Typical RAG | NotebookLM | Notion AI | | ------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------ | --------- | ----------- | ---------- | --------- | | Weighted knowledge graph + D3 visualisation — [[wikilinks]] auto-built at ingest; co-source edges connect pages compiled from the same document; edge thickness reflects combined weight; dashed edges for co-source-only relationships; Louvain cluster colouring; click node to query; Obsidian graph panel: same graph inside Obsidian via Canvas, type filter, click node to open page | Yes | No | Partial | No | | Orphan page detection — unreferenced pages surfaced by lint with ready-to-paste index entries | Yes | No | No | No | | Query-scoped routing — ROUTING.md maps wiki branches to page slugs; queries auto-select relevant branches; new pages auto-slotted | Yes | No | No | No | | Candidates staging — ingest pages to a staging area first; review, promote, or discard before they enter the live wiki | Yes | No | No | No | | Scaffold automation — regenerates index categories, AGENTS.md/CLAUDE.md/GEMINI.md, and purpose.md from current wiki state; protected pages never overwritten | Yes | No | No | No | | Domain templates — 30 fully-authored templates (finance, technology, healthcare, legal, and more) so you start with the right agent guidelines, routing, and structure from day one | Yes | No | No | No |

Search & Query

| Capability | Synthadoc | Typical RAG | NotebookLM | Notion AI | | --------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- | ------------------ | ----------- | ---------- | --------- | | Query decomposition + gap detection — compound questions split into parallel BM25 sub-queries; thin results trigger a knowledge-gap callout with suggested web searches | Yes | Partial | No | No | | BM25 TF fallback + compound identifier search — reliable results on small corpora (IDF collapse → TF fallback); underscore identifiers expanded at index and query time so capex growth matches capex_growth | Yes | No | No | No | | Web search → wiki pages — Tavily search fans out into parallel URL ingest jobs; gap callout in web UI suggests searches inline | Yes | No | No | No | | Semantic re-ranking — optional vector re-ranking (BAAI/bge-small-en-v1.5) improves recall on conceptually related queries; BM25 stays as fallback | Yes (optional) | Varies | No | No | | Streaming output + query cache — token-by-token streaming; cache key = question + wiki version; auto-invalidates on ingest or lifecycle change | Yes | Partial | Partial | Partial | | Proportional context budget — sources allocated proportionally to model context window (60 % wiki / 20 % history / 15 % system / 5 % index); replaces fixed top-N cap | Yes | No | No | No |

Interfaces & Integration

| Capability