Subscribe a book from the marketplace, upload your own docs, or let your agent write what it learns — your agents cite every answer by page, not by vibes.
curl -fsSL https://getcandlekeep.com/install.sh | shInstalls, authenticates, and connects your agents.
For developers building with Claude Code, Cursor, or Codex.
Free forever tier · No credit card · One command to install
CandleKeep is a library your AI agents read while they work — they open a book, find the page, and cite it.
No vector search. Agents read pages directly, like a person reading a book — and cite them.
Books are structured into pages and tables of contents so agents can actually read them.
It's a library of outside expertise for your agent, not an internal wiki to govern.
No vector databases. No RAG pipelines. Your agent checks the table of contents, picks the right chapter, and reads the relevant pages.
One command installs CandleKeep. Your agent detects research questions automatically — no manual references, no configuration per conversation.
Your agent reads like a developer explores a codebase. Table of contents first, then relevant chapters, then specific pages. No keyword search. No vector similarity. Actual reading.
Training data is like remembering a book you read 3 years ago. CandleKeep is like having it open on your desk. Every answer includes the source and page number.
Define a knowledge area. Your agent proposes additions from every coding session. Over time, it builds a structured book from your real experience.
Topics: race conditions · async · deadlocks
Chapters written by your agent:
Built from 47 coding sessions
Define topics and criteria. Your agent handles the rest.
Knowledge accumulates automatically from your real work.
Who it's for
However you build with AI, CandleKeep gives your agents the knowledge they were missing.
This is you if… You build agents, tools, MCP servers, or skills in Claude Code, Cursor, or Codex.
Inject the domain knowledge your agent's training data doesn't have.
This is you if… You're shipping with AI without a deep background in the domain.
Hand your agent the right book and it instantly works like a specialist.
This is you if… Your team runs many agents that all start from zero.
Give every agent the same curated, cited knowledge — and let them write back what they learn.
CandleKeep integrates with the agents and platforms you already use.
Coding Agents
Platforms & Interfaces
Your agent handles the code. CandleKeep gives it the domain knowledge.
Your agent reads the security textbook so it audits code like a specialist, not a generalist.
See how →Your agent synthesizes insights from 12 business books to craft strategy, not generic advice.
See how →Your agent applies design principles from 7 books, not generic 'make it pretty' advice.
See how →Deep dives into building knowledge libraries for AI agents.
An agent tested our Claude Code plugin's startup hook and reported a clean pass. On the second pass a different agent wrote: "the evidence I reported proves nothing." It was right. How we built a test container that holds the coding agent, proves the hook fired, and installs the product on three operating systems as a stranger, and the book that came out of it.
Read moreA user asked his coding agent for a security review, the librarian found exactly the right book for free on the marketplace, and the readers came back with "Item not found." Fixing that meant answering a harder question: when should an agent decide on its own to stop working from memory and go look something up? An engineering account of the trigger we built, the three instruction wordings that failed, and what the trials changed.
Read moreAn LLM wiki is a knowledge base your AI agent writes and maintains itself — so your knowledge compounds across sessions instead of being re-derived every time. If you use Claude Code seriously, you've probably already hacked one together. Here's the version that took me an afternoon to build — because it was just a book.
Read morePosts, threads, and videos about CandleKeep from across the web.
The first library built for agents — in an era where everyone has access to powerful models, the model's knowledge becomes the baseline and the competitive edge shifts to the layer below it: the data and context the model can see.
Curated knowledge libraries any AI agent can pull from on demand — web security, design, product strategy, and more.
Agent-optimized webapp security reference
An architectural analysis of Claude Code -- the internal agent schema, built-in agents, coordinator system, hooks, memory, and token-optimization decisions. Written from source analysis, in our own words.
A 20,000-word, 12-chapter technical book by Sahar Carmel about building multi-agent AI systems. Covers architecture, memory systems, knowledge systems (books over RAG), inter-agent communication, security, deployment, and development workflows.
A comprehensive reference manual for AI agents building web and mobile interfaces. 15 self-contained chapters covering visual hierarchy, typography, color systems, spacing, component patterns, navigation, cognitive load, behavioral triggers, persuasion, form design, responsive design, accessibility, and anti-patterns. Synthesized from 10 foundational UI/UX books (Refactoring UI, Don't Make Me Think, Laws of UX, 100 Things Every Designer, Hooked, Influence) into 170+ concrete, actionable rules with specific thresholds, values, and anti-patterns. Each rule uses When/Do/Values/Don't format for direct agent application.
by CandleKeep
A comprehensive guide covering skill structure, SKILL.md writing, progressive disclosure, examples, and best practices for building Claude Code skills.
Complete reference for AI coding agents performing code reviews. 18 chapters + appendices, 231 rules with concrete thresholds, TypeScript and Python examples, machine-checkable verification criteria, and book-grounded citations. Covers naming, functions, classes, code smells, type safety, error handling, performance, testing, concurrency, security, API design, dependencies, database/ORM patterns, observability, and diff-scanning heuristics.
You've seen knowledge tools before. Here's why CandleKeep isn't one of them.
vs NotebookLM
NotebookLM helps humans talk to docs.
CandleKeep
CandleKeep helps AI agents read books while they work.
vs DIY RAG
Stop building pipelines.
CandleKeep
Install a book in one command.
vs CLAUDE.md / Context
CLAUDE.md tells your agent what to do.
CandleKeep
CandleKeep teaches it what to know.
vs Code Docs Tools
Code docs describe your codebase.
CandleKeep
CandleKeep gives your agent domain expertise.
Everything you need to know before getting started.
One command installs the CLI, authenticates, and connects your agents.
curl -fsSL https://getcandlekeep.com/install.sh | shInstalls the CLI, authenticates, and configures Claude Code, Codex, and Cursor.
Scan this codebase and set up my CandleKeep library for this project. 1) Analyze the repo — languages, frameworks, architecture, and domain areas. 2) Browse the marketplace and subscribe to the top 3–5 most relevant books. 3) Explain why each book helps this codebase. 4) Suggest documents I should upload from my own knowledge.
Paste it into Claude Code, Cursor, or Codex after installing — your agent curates its own starting library.
Start free. Upgrade when you need more.
Free forever · no card required
A focused library for the topics you read most.
Less than a coffee a day · cancel anytime
For power users who want to publish and let their agent write.
Up to 5 seats · $19.80 / user / month
Shared libraries and centralized billing for teams.