llm-wiki

A shared human and AI memory, and the ecosystem it needs

Chris Sweet
Charles Vardeman
Priscila Moreira
James Sweet
Sam Njoroge

November 12, 2026

What llm-wiki is

A compounding knowledge base, written by the agent and ruled by the human

Human and AI agent either side of a shared wiki

Where llm-wiki came from

Five steps, four people, one lesson

Timeline of the project's origin

Installation is now easy

Everything from inside the Codex or Claude app: marketplace, plugin, skills, hooks

Plugins screen, Add a marketplace
1. Add the marketplace. Plugins, Add, then paste the llm-wiki-colab repository address.
llm-wiki plugin page with Install plugin
2. Install the plugin. Open llm-wiki under Personal and choose Install plugin.
Installed plugin with per-skill switches
3. Pick the features. Each skill has its own switch: wiki-init, wiki-source, wiki-experiment, wiki-lint, wiki-ask, wiki-enroll.
Hooks settings with SessionStart and PostToolUse enabled
4. Check the hooks. SessionStart loads the index and recent log; PostToolUse runs the verification gate.

What the plugin gives you

Core functionality, plus helpers

Anatomy of a wiki page with index, log, schema, helpers and hooks

Does the memory work?

Forty questions on one real research project

Bar chart: 2 percent, 42 percent, 90 percent correct

What we hoped for

The vision: a fabric that connects wikis, data and marketplaces

The fabric at the center, with wikis, data and marketplaces around it

Why this matters for lab automation

Without the ecosystem, every instrument needs a hand-written adapter

Pipeline from instrument to wiki with the adapter step missing

The ask

For each thing you build or run, three questions

Three questions: a skill, a plugin, a fabric source