# nashsu/llm_wiki

Source: [GitHub](https://github.com/nashsu/llm_wiki)  
Feed7 permalink: https://feed7.dev/p/llm-wiki-1dqmknq  
Published: Unknown  
Trust: Needs Review (needs_review)

## Why Included

LLM Wiki compiles source files into a persistent, traceable wiki that agents can query through local APIs, MCP, and skills, reducing repeated retrieval and re-analysis.

## Source Summary

LLM Wiki turns imported material into a persistent three-layer system: **raw sources → generated wiki → schema**. Its two-call ingest separates analysis from page generation, records source links, skips unchanged files with a SHA-256 cache, and recovers a persistent queue after restarts.

## Practical Implication

For agent context, treat the maintained wiki as a reusable compiled layer rather than rebuilding answers from raw documents every turn. The local API, **MCP server**, and agent skill expose hybrid search, file reads, graph traversal, and rescans to Claude Code or Codex.

## Agent-Ready Context

LLM Wiki turns imported material into a persistent three-layer system: **raw sources → generated wiki → schema**. Its two-call ingest separates analysis from page generation, records source links, skips unchanged files with a SHA-256 cache, and recovers a persistent queue after restarts.

For agent context, treat the maintained wiki as a reusable compiled layer rather than rebuilding answers from raw documents every turn. The local API, **MCP server**, and agent skill expose hybrid search, file reads, graph traversal, and rescans to Claude Code or Codex.

Vector retrieval is optional and reportedly raises recall from **58.2% to 71.4%**, but the material does not describe the benchmark dataset or evaluation method. Generated pages can still distort sources, so source-only answering, traceability, linting, and human review remain important.

## Connected Context

Feed7 judgment across 757 accumulated Signals:

This makes external agent memory concrete as a compiled, inspectable knowledge layer rather than learned or repeatedly re-ingested context. It reinforces selective retrieval and structured navigation, while narrowing trust: generated wiki pages need source links, source-only fallback, and review because persistence can preserve distortions as readily as useful synthesis. The reported recall gain remains directional without evaluation details.

- [thedotmack/claude-mem](https://feed7.dev/p/claude-mem-1uba088) — Both keep persistent memory external and selectively retrievable; LLM Wiki adds a maintained schema and source-linked wiki layer, while leaving similar governance and contradiction risks.
- [Citation Needed: Provenance for LLM-Built Knowledge Graphs — Daniel Chalef, Zep AI](https://feed7.dev/p/citation-needed-provenance-for-llm-built-knowledge-graphs-daniel-chalef-1iob5t8) — The provenance warning directly qualifies LLM Wiki’s generated layer: simple source links may be insufficient when a page synthesizes, changes, or merges claims.
- [Scaling Compute on Context — Jack Morris, Engram](https://feed7.dev/p/scaling-compute-on-context-jack-morris-engram-1pxr9bt) — LLM Wiki supplies a retrieval-and-compression alternative to training on a private corpus, consistent with the finding that corpus memorization alone does not make knowledge usable.
- [SheetCompass: Hierarchical Relation Graphs for Agentic Spreadsheet Reasoning](https://feed7.dev/p/2608-14452v1-17xm2vb) — Both preserve structure instead of flattening source material, but SheetCompass specializes that principle for spreadsheet relations while LLM Wiki applies a general wiki and schema.

## Context Map

- Layer: context
- Domains: research, data
- Topics: retrieval, context-engineering, agent-memory

## Uncertainty

- Vector retrieval is optional and reportedly raises recall from **58.2% to 71.4%**, but the material does not describe the benchmark dataset or evaluation method. Generated pages can still distort sources, so source-only answering, traceability, linting, and human review remain important.

## Agent Instruction

Use this item as source-backed context. Do not invent claims beyond the linked source. If this item conflicts with another source, call out the conflict.