The LLM wiki you've been hacking together — without the hacking
An LLM wiki is a knowledge base your AI agent writes and maintains itself. CandleKeep gives you one out of the box — seeded the moment you install, the least complicated manuscript in your library.
You already know the pattern. You've probably built it.
- Every new session, your agent forgets everything — the decisions you made, the gotchas you found, the conventions you settled on.
- So you built a wiki for it: a raw folder, a schema file, a search script, maybe Obsidian. And now you're the janitor of your own memory system.
- It drifts. The scripts break. It never quite feels like a real system — and the maintenance is on you.
In CandleKeep, the wiki is just a book
That's the whole trick. A table of contents, a changelog, version history, links between pages — CandleKeep already had all of it. So the LLM wiki isn't a new system to learn. It's the simplest manuscript you own: zero config, seeded for you, written entirely by your agent.
Table of contents
Auto-built from the book's headings.
Changelog + versions
Every edit dated and reversible.
Page interlinks
Pages point to each other; the agent follows them.
Read like a book
Browsed and cited by page — not chunked.
Three steps. Two of them are nothing.
- 1
Install CandleKeep
One command. Your library and your LLM wiki are created together.
- 2
Your wiki is seeded and ready
A "My LLM Wiki" book is waiting — no schema to write, no directory to maintain.
- 3
Your agent reads and maintains it — every session
It files concepts, decisions, and patterns automatically, and reads them back next time.
The DIY LLM wiki vs. CandleKeep
Build it yourself
- A raw/ directory you keep feeding
- A hand-written CLAUDE.md schema
- A search script (or a vector DB) to query it
- An Obsidian vault to view and link it
- And the ongoing job of keeping it all from drifting
CandleKeep
- Seeded for you on install — zero config
- Your agent writes every page, you just work
- Table of contents + changelog, maintained automatically
- Read by your agent on every session, with citations
- Versioned — every change is reversible
Questions
- What is an LLM wiki?
- An LLM wiki is a structured, interlinked knowledge base that an AI agent writes and maintains itself — summaries, concept pages, an index, and a changelog — so your knowledge compounds across sessions instead of being re-derived each time. In CandleKeep it's a self-maintaining book your coding agent updates after each session.
- What does "seeded on install" mean?
- Every CandleKeep library comes with a pre-created "My LLM Wiki" book the moment you sign in. There's nothing to scaffold — no raw/ directory, no CLAUDE.md schema, no search engine to wire up. Your agent starts maintaining it on the next session.
- How does the agent read it?
- Like a book, not a vector database. The agent reads the table of contents, then the pages it needs, and cites them — no embeddings, no chunking. New entries are appended with an index and a dated changelog so the next session can navigate what came before.
- Is this RAG?
- No. RAG retrieves chunks at query time and rebuilds context from scratch every question. An LLM wiki is the opposite: the knowledge is compiled once into pages the agent reads directly, and kept current as you work. Full-document reading with citations, not similarity search.
Want the longer story of how it was built? Read the build post →
You already have an LLM wiki. Go meet it.
Install CandleKeep and your agent starts compounding your knowledge from the very next session. Free to start, no credit card.
Install CandleKeep