geronimo-iia/llm-wiki

geronimo-iia/llm-wiki (13 stars, Apache 2.0) is a headless wiki engine with 23 MCP tools in a single Rust binary. No LLM inside — pure infrastructure that manages files, git, full-text search, and graph structure.

Architecture (confidence: high)

Implements the Dynamic Knowledge Repository (DKR) pattern: process sources at ingest time, not query time. The engine validates frontmatter against JSON Schema, extracts typed graph edges from sources and concepts fields, and indexes everything in Tantivy. The graph is live the moment a page is committed.

Component	Technology
Search	tantivy — BM25
Git	git2 — libgit2 bindings
Graph	petgraph — typed DiGraph
MCP	rmcp — stdio + Streamable HTTP
ACP	Agent Client Protocol — session-oriented streaming

Features

Core Architecture

• Headless engine — no LLM inside, pure infrastructure
• Single Rust binary — no runtime, no database, no Docker
• DKR pattern — process at ingest time, not query time
• Three components — engine (Rust) + skills (Markdown) + Hugo CMS

Ingest & Processing

• wiki_ingest — validate + index + commit in one operation
• JSON Schema validation — frontmatter validated against schema
• Typed edge extraction — extracts graph edges from sources/concepts fields
• Tantivy indexing — BM25 full-text search index

Search & Retrieval

• wiki_search — BM25 faceted search
• wiki_suggest — by tag overlap, graph distance, BM25
• wiki_list — paginated listing

Knowledge Graph

• petgraph DiGraph — typed directed graph
• wiki_graph — Mermaid/DOT/llms export formats
• wiki_stats — graph statistics

MCP / Tools

• 23 MCP tools — most comprehensive MCP-tooled wiki engine
• ACP support — Agent Client Protocol for IDE integration
• wiki_export — llms.txt export for agent consumption
• wiki_lint — deterministic quality rules

Quality & Governance

• wiki_lint — deterministic rules, no LLM needed
• wiki_schema — schema introspection
• wiki_history — full audit trail

Integration

• Claude Code skills — llm-wiki-skills plugin
• Hugo CMS — llm-wiki-hugo-cms for website rendering
• MCP + ACP — dual protocol support
• Mermaid/DOT — diagram export

Data & Storage

• Markdown files — source of truth
• Git-backed — libgit2 bindings
• Tantivy index — local full-text search
• No database — pure file-based

23 MCP Tools (confidence: high)

Core tools include: wiki_search (BM25 faceted), wiki_list (paginated), wiki_content_read/write/new, wiki_resolve (slug or wiki:// URI), wiki_ingest (validate + index + commit), wiki_graph (Mermaid/DOT/llms), wiki_suggest (by tag overlap, graph distance, BM25), wiki_stats, wiki_lint (deterministic rules), wiki_export (llms.txt), wiki_history, wiki_schema, wiki_spaces_*

Three Components (confidence: high)

• llm-wiki-engine — pure Rust binary
• llm-wiki-skills — Claude Code plugin with workflow skills
• llm-wiki-hugo-cms — Hugo scaffold to render wiki as website

Design Philosophy (confidence: high)

• No opinions about workflows, LLM providers, or interfaces
• Every LLM call happens outside the binary
• Every workflow lives in a skill (Markdown)
• Markdown files + git are the source of truth
• Single binary, no runtime, no database, no Docker

Implications

This is the most comprehensive MCP-tooled wiki engine in the ecosystem — 23 tools vs krakiun/llmwiki’s 5-9. The separation of engine (Rust binary) from skills (Markdown) from rendering (Hugo) is the cleanest three-component architecture seen. The ACP (Agent Client Protocol) support for IDE integration is unique. Like krakiun/llmwiki, it solves the scaling problem by replacing index.md with a search index, but adds graph, suggestions, and validation on top.

Related

• krakiun-llmwiki — Smaller scope, same stack (Rust + MCP)
• llm-wiki — The pattern this engines
• llm-wiki-mcp — Minimal 4-tool MCP server (much simpler)
• model-context-protocol — MCP is the protocol this uses

Sources

• raw/articles/geronimo-iia-llm-wiki-engine.md