📑 What is OpenKB

OpenKB (Open Knowledge Base) is an open-source system (in CLI) that compiles raw documents into a structured, interlinked wiki-style knowledge base using LLMs, powered by PageIndex for vectorless long document retrieval.

The idea is based on a concept described by Andrej Karpathy: LLMs generate summaries, concept pages, and cross-references, all maintained automatically. Knowledge compounds over time instead of being re-derived on every query.

Why not traditional RAG?

Traditional RAG rediscovers knowledge from scratch on every query. Nothing accumulates. OpenKB compiles knowledge once into a persistent wiki, then keeps it current. Cross-references already exist. Contradictions are flagged. Synthesis reflects everything consumed.

OpenKB has two layers: a wiki foundation that compiles and maintains your knowledge, and generators (query / chat / Skill Factory) that turn it into useful output. See Usage for the full command list.

🚀 Getting Started

Install

Quick Start

# 1. Create a directory for your knowledge base
mkdir my-kb && cd my-kb

# 2. Initialize the knowledge base
openkb init

# 3. Add documents
openkb add paper.pdf
openkb add ~/papers/                            # Add a whole directory
openkb add https://arxiv.org/pdf/2509.11420     # Or fetch from a URL

# 4. Ask a question
openkb query "What are the main findings?"

# 5. Or chat interactively
openkb chat

# 6. Or distill your wiki into a redistributable skill
openkb skill new my-expert "Reason like an expert on <topic-from-your-docs>"

Set up your LLM

OpenKB comes with multi-LLM support (e.g., OpenAI, Claude, Gemini) via LiteLLM (pinned to a safe version).

Set your model during openkb init, or in .openkb/config.yaml, using provider/model LiteLLM format (like anthropic/claude-sonnet-4-6). OpenAI models can omit the prefix (like gpt-5.4).

Create a .env file with your LLM API key:

LLM_API_KEY=your_llm_api_key

🧩 How OpenKB Works

Architecture

raw/                              You drop files here
 │
 ├─ Short docs ──→ markitdown ──→ LLM reads full text
 │                                     │
 ├─ Long PDFs ──→ PageIndex ────→ LLM reads document trees
 │                                     │
 │                                     ▼
 │                         Wiki Compilation (using LLM)
 │                                     │
 ▼                                     ▼
wiki/                                  │            ← the foundation
 ├── index.md            Knowledge base overview
 ├── log.md              Operations timeline
 ├── AGENTS.md           Wiki schema (LLM instructions)
 ├── sources/            Full-text conversions
 ├── summaries/          Per-document summaries
 ├── concepts/           Cross-document synthesis ← the good stuff
 ├── entities/           Specific named things (people, orgs, places, products)
 ├── explorations/       Saved query results
 └── reports/            Lint reports
                                       │
                ┌──────────────────────┼──────────────────────┐
                ▼                      ▼                      ▼
            query / chat         Skill Factory          (future)
          (LLM answers from     openkb skill new       ppt / podcast /
            the wiki)           → output/skills/        report / …
                                + marketplace.json

Short vs. Long Document Handling

Short docs are read in full by the LLM. Long PDFs are indexed by PageIndex into a hierarchical tree with summaries. The LLM reads the tree instead of the full text, enabling better retrieval from long documents.

Knowledge Compilation

When you add a document, the LLM:

1. Generates a summary page
2. Reads existing concept and entity pages
3. Creates or updates concepts with cross-document synthesis
4. Creates or updates entity pages (people, orgs, places, products)
5. Updates the index and log

A single source might touch 10-15 wiki pages. Knowledge accumulates: each document enriches the existing wiki rather than sitting in isolation.

⚙️ Usage

OpenKB commands fall into two layers: the wiki foundation (compile + manage your knowledge) and generators (turn that wiki into useful output).

🧱 Wiki Foundation — compile and maintain

✨ Generators — turn the wiki into output

A "generator" reads from the compiled wiki and produces something usable: an answer, a conversation, a skill folder. The wiki is the substrate; generators are the surfaces.

Query & Chat — ask the wiki

openkb query "..." answers a single question. openkb chat is interactive — each turn carries history, so you can dig into a topic without re-typing context. Both use the same underlying wiki and the same retrieval primitives (PageIndex for long docs, direct concept reads for short).

openkb query "What does the literature say about attention scaling?"

openkb chat                       # start a new session
openkb chat --resume              # resume the most recent session
openkb chat --resume 20260411     # resume by id (unique prefix works)
openkb chat --list                # list all sessions
openkb chat --delete <id>         # delete a session

Inside a chat, type / to access slash commands (Tab to complete):

• /help — list available commands
• /status — show knowledge base status
• /list — list all documents
• /add <path> — add a document or directory without leaving the chat
• /skill new <name> "<intent>" — compile a skill from this chat (see below)
• /save [name] — export the transcript to wiki/explorations/
• /clear — start a fresh session (the current one stays on disk)
• /lint — run knowledge base lint
• /exit — exit (Ctrl-D also works)

🛠 Skill Factory — Drop in a book. Out comes a digital expert.

The newest generator. openkb skill new distills any subset of your wiki into an Anthropic Skill — a portable folder that Claude Code, Codex CLI, Gemini CLI, and Cursor all install and load natively. Drop in a book's worth of papers; out comes a specialist that other agents can call on.

openkb skill new karpathy-thinking \
  "Reason about transformers and attention in Karpathy's style"

This produces:

<kb>/output/skills/karpathy-thinking/
├── SKILL.md                   # YAML frontmatter + when-to-use + approach
├── references/                # depth material the agent loads on demand
│   ├── methodology.md
│   └── key-quotes.md
└── (scripts/)                 # optional, only if intent implies computation

…plus an auto-updated <kb>/.claude-plugin/marketplace.json so the whole KB is one-line installable.

Install locally:

cp -r output/skills/karpathy-thinking ~/.claude/skills/

Share with others — push your KB to GitHub, then anyone runs:

npx skills@latest add <your-org>/<your-repo>

Iterate from chat — compilation is one-shot, but follow-up edits aren't. Inside openkb chat, you can refine without re-running the whole pipeline:

/skill new karpathy-thinking "Reason about transformers like Karpathy"
[generation streams]
> description is too generic, make it about transformer implementations specifically
[agent edits SKILL.md frontmatter in place]

Quality gates — structural validation, trigger-accuracy + body-coverage evaluation, and full history/rollback:

# Lint structure (auto-runs at end of `skill new`)
openkb skill validate karpathy-thinking
openkb skill validate --strict          # treat warnings as failures

# Does the description actually fire when it should?
openkb skill eval karpathy-thinking --save

# History + rollback if a new iteration regresses
openkb skill history karpathy-thinking
openkb skill rollback karpathy-thinking --to 2

Configuration

Settings are initialized by openkb init, and stored in .openkb/config.yaml:

model: gpt-5.4                   # LLM model (any LiteLLM-supported provider)
language: en                     # Wiki output language
pageindex_threshold: 20          # PDF pages threshold for PageIndex

entity_types (optional): a YAML list overriding the entity-type vocabulary used for entity pages; omit it to use the default person, organization, place, product, work, event, other.

extra_headers (optional): a YAML mapping of extra HTTP headers sent with every LLM request (forwarded to LiteLLM's extra_headers). Useful for providers that expect custom headers, e.g. GitHub Copilot IDE-auth headers:

extra_headers:
  Editor-Version: vscode/1.95.0
  Copilot-Integration-Id: vscode-chat

Subscription-based providers that authenticate via OAuth device flow (e.g. chatgpt/*, github_copilot/*) need no API key — OpenKB skips the missing-key warning for them.

Model names use provider/model LiteLLM format (OpenAI models can omit the prefix):

PageIndex Integration

Long documents are challenging for LLMs due to context limits, context rot, and summarization loss. PageIndex solves this with vectorless, reasoning-based retrieval — building a hierarchical tree index that lets LLMs reason over the index for context-aware retrieval.

PageIndex runs locally by default using the open-source version, with no external dependencies required.

Optional: Cloud Support

For large or complex PDFs, PageIndex Cloud can be used to access additional capabilities, including:

• OCR support for scanned PDFs (via hosted VLM models)
• Faster structure generation
• Scalable indexing for large documents

Set PAGEINDEX_API_KEY in your .env to enable cloud features:

PAGEINDEX_API_KEY=your_pageindex_api_key

AGENTS.md

The wiki/AGENTS.md file defines wiki structure and conventions. It's the LLM's instruction manual for maintaining the wiki. Customize it to change how your wiki is organized.

At runtime, the LLM reads AGENTS.md from disk, so your edits take effect immediately.

Using with Obsidian

OpenKB's wiki is a directory of Markdown files with [[wikilinks]]. Obsidian renders it natively.

1. Open wiki/ as an Obsidian vault
2. Browse summaries, concepts, and explorations
3. Use graph view to see knowledge connections
4. Use Obsidian Web Clipper to add web articles to raw/

Using with Claude Code / Codex / Gemini CLI

OpenKB ships a SKILL.md so any agent CLI can read your compiled wiki — no extra runtime, no MCP setup, just install the skill once.

Claude Code:

/plugin marketplace add VectifyAI/OpenKB
/plugin install openkb@vectify

Gemini CLI:

gemini skills install https://github.com/VectifyAI/OpenKB.git --path skills/openkb --consent

OpenAI Codex CLI (no marketplace command yet — manual symlink):

git clone https://github.com/VectifyAI/OpenKB.git ~/openkb-src
mkdir -p ~/.agents/skills
ln -s ~/openkb-src/skills/openkb ~/.agents/skills/openkb

The skill is read-only — it won't run openkb add, remove, or lint --fix without you asking. See skills/openkb/SKILL.md for the full instruction set.

🧭 Learn More

Compared to Karpathy's Approach

The Stack

• PageIndex — Vectorless, reasoning-based document indexing and retrieval
• markitdown — Universal file-to-markdown conversion
• OpenAI Agents SDK — Agent framework (supports non-OpenAI models via LiteLLM)
• LiteLLM — Multi-provider LLM gateway
• Click — CLI framework
• watchdog — Filesystem monitoring

Roadmap

• Extend long document handling to non-PDF formats
• Scale to large document collections with nested folder support
• Hierarchical concept (topic) indexing for massive knowledge bases
• Database-backed storage engine
• Web UI for browsing and managing wikis

Contributing

Contributions are welcome! Please submit a pull request, or open an issue for bugs or feature requests. For larger changes, consider opening an issue first to discuss the approach.

License

Apache 2.0. See LICENSE.

Support Us

If you find OpenKB useful, please give us a star 🌟 — and check out PageIndex too!