# 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-

- **URL**: https://www.freshcrate.ai/projects/synthadoc
- **Author**: axoviq-ai
- **Category**: AI Agents
- **Latest version**: `v0.7.0` (2026-06-05)
- **License**: AGPL-3.0
- **Source**: https://github.com/axoviq-ai/synthadoc
- **Language**: Python
- **GitHub**: 66 stars, 10 forks
- **Registry**: github (`axoviq-ai/synthadoc`)
- **Tags**: `agent-skills`, `agentic-ai`, `cli-tool`, `domain-adaptation`, `enterprise`, `enterprise-solutions`, `knowledge-graph`, `local-llm`, `python`

## Description

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.

## Recent releases

| Version | Date | Urgency | Changes |
| --- | --- | --- | --- |
| `v0.7.0` | 2026-06-05 | High | ## Synthadoc v0.7.0 Community Edition: Key Highlights  ##  Local Web Chat UI  'synthadoc web' CLI starts a local chat interface in your browser, local-only, nothing leaves your machine. Use it the same way you use the CLI: ask questions about your wiki's domain and get streamed, cited answers.   Beyond content queries, the UI also auto-detects the health of your wiki (Explorer, Health Check, or Power User mode) and surfaces context-appropriate prompts: if your wiki has orphan pages or cont |
| `v0.6.0` | 2026-05-29 | High | ## Synthadoc v0.6.0 Community Edition: Key Highlights  # Page lifecycle machine  5-State Page Lifecycle: pages now move through five states: draft → active → stale → contradicted → archived. Transitions are mostly automatic: a clean lint pass promotes draft to active; a SHA-256 hash mismatch on the source file marks the page stale; ingesting a conflicting source marks it contradicted. Every transition is permanently logged with who triggered it, when, and why, so you have an auditable trail |
| `v0.5.0` | 2026-05-22 | High | ## Synthadoc v0.5.0 Community Edition:  Key Highlights  ### Adversarial Review   A second independent LLM interrogates every compiled page for epistemic overreach, unsupported generalisations, and internal contradictions - using a different model from the one that wrote the page. Warnings land in a dedicated tab in the Obsidian lint modal, in YAML frontmatter (lint_warnings) on each page, and in the lint report JSON. Assign any provider/model as the judge (agents.adversarial in config.toml) |
| `v0.4.0` | 2026-05-11 | High | # Synthadoc v0.4.0 Feature Highlights  v0.4.0 addresses what happens when the wiki grows large enough that a flat architecture starts showing cracks - query scope, write-path quality, and piping structured knowledge into external agents without losing control of the token envelope.  - Routing layer with branch taxonomy: A ROUTING.md file at the wiki root maps topic branches to page slugs, using the same ## Heading → [[slug]] structure as index.md. At query time, an LLM selects the 1–2 most r |
| `v0.3.0` | 2026-05-04 | High | ## Synthadoc v0.3.0 Feature Highlights  v0.3.0 expands on the same architecture - the big additions this cycle are around what can go into the wiki, and removing the friction of needing a separate API key.   ( https://github.com/axoviq-ai/synthadoc )  - Zero-API-key LLM providers - if you already pay for Claude Code or Opencode, one config line (provider = "claude-code") routes all three agents (ingest, query, lint) through your existing subscription. No separate API keys or Anthropic/OpenA |
| `v0.2.0` | 2026-04-24 | High | ## Synthadoc v0.2.0 Feature Highlights:  - Query decomposition: compound questions automatically split into parallel BM25 sub-queries and merged before synthesis - Knowledge gap detection: queries that hit thin coverage return targeted synthadoc ingest "search for: …" suggestions as an Obsidian callout - Web search decomposition: broad search intents split into focused keyword queries before Tavily fetch, with URL deduplication and a configurable result cap - Semantic re-ranking: opt-in vec |
| `v0.1.0` | 2026-04-15 | High | ## What's Changed * Release v0.1 — skills v2, Obsidian plugin enhancements, bug fixes by @paulmchen in https://github.com/axoviq-ai/synthadoc/pull/1 * v0.1 complete — startup banner, doc updates, pyproject cleanup by @paulmchen in https://github.com/axoviq-ai/synthadoc/pull/2 * feat: v0.1 — multi-provider LLM, web search, two-step ingest, purpose/overview pages, audit CLI by @paulmchen in https://github.com/axoviq-ai/synthadoc/pull/3 * fix: add missing ingest_agent.py changes and update docs |

## Citation

- HTML: https://www.freshcrate.ai/projects/synthadoc
- Markdown: https://www.freshcrate.ai/projects/synthadoc.md
- Dependencies JSON: https://www.freshcrate.ai/api/projects/synthadoc/deps

_Generated by freshcrate.ai. Indexes github releases for AI-agent ecosystem packages._