Browse / LLM wiki / llm-wiki

llm-wiki

Uraxii

An LLM wiki: a live model of an environment plus the accumulated reasoning over it.

Open Uraxii/llm-wiki → Checked against the four rules.

Works with any agent

ships AGENTS.md, the cross-tool instruction file.

In its own words

README.mdUraxii/llm-wiki

An LLM wiki: a live model of an environment plus the accumulated reasoning over it. Built to give agents cheap, easy-to-retrieve context.

What it is

Three layers, no service, no database, no network surface.

  • **sources/.** Immutable. Articles, papers, snapshots. Written once, never

edited, never deleted.

  • **wiki/.** The model owns it. Pages it writes and keeps updating as new

sources land, linked with [[wikilink]] syntax. wiki/index.md is generated.

  • **SCHEMA.md.** The configuration that matters: conventions, workflows, and

which retrieval path to take for which question.

Retrieval is by meaning: pages are embedded and search ranks them by cosine similarity. Grep still works and is still the right tool for an exact string. See docs/design/llm-wiki.md for the full design and skills/llm-wiki/SKILL.md for how an agent uses it.

CLI

Install: uv tool install ./skills/llm-wiki from the repo root. Needs Python 3.14 and sqlite-vec, the project's one dependency.

llmwiki verbs: init, where, ingest, summarize, dedup, lint, embed, status, search.

Sources, summary pages, story pages, log.md and the vector store are the CLI's. Everything else under wiki/, index.md included, belongs to the agent, and the CLI never overwrites it.

Quoting 53% of the README. Read the rest on GitHub ↗

Read from the repository's git tree on 2026-08-17.