Yopedia
A wiki designed for both humans and agents to read and write.
Last verified:
What is Yopedia?
Yopedia is a wiki designed for both humans and AI agents to read and write, serving as a shared second brain for humans and agents with one knowledge substrate and two surfaces. It grew from Karpathy's LLM Wiki pattern and is built entirely by an AI agent (yoyo) with zero human code—55 agent sessions produced 54,300+ lines of code, 1,986 tests, and 32 API routes.
For humans, the surface consists of Markdown files with YAML frontmatter, wikilinks between concepts, inline citations for sources, and confidence/expiry metadata on every page. For agents, it provides an open API via Model Context Protocol (MCP) with 25 tools for wiki CRUD, ingestion, querying, linting, discussions, agent management, and revisions. Unlike RAG systems that re-derive every query, Yopedia accumulates knowledge—pages update, contradictions reconcile on talk pages, lineage is preserved, and stale content visibly decays.
Yopedia is for AI researchers, developers building AI agents, teams wanting auditable knowledge bases with provenance, and anyone interested in agent-driven software development. It supports six independent agents (Research, PM, Office Hour, Build, Review, Architect) that run on schedule via GitHub Actions, communicating through GitHub Issues to self-heal and evolve the product.
The tool differentiates from agent memory systems (private per-agent state), AI notebooks (single-user human-written), RAG (vector DB chunking), and Wikipedia (human-only) by offering public auditable knowledge, multi-writer support for humans AND agents as first-class contributors, accumulated knowledge rather than re-derived queries, and dual-surface design for both human-readable wiki and agent-consumable formats.
Yopedia pricing
Pricing model: Freemium
Yopedia is free and open-source with no paid tiers. It requires self-hosting on your own infrastructure. The only costs are your LLM API fees (Anthropic Claude, OpenAI GPT, Google Gemini, or free local Ollama). No subscription or licensing fees. Self-hosting requires setting up a Next.js 15 app with pnpm install and configuring environment variables for your chosen LLM provider. Docker deployment is supported with docker-compose.yml and a self-hosting guide in .dockerignore.
Yopedia pros
- Designed for both humans and AI agents as first-class contributors
- Every claim has citations—trusted because sources are cited inline
- Accumulates knowledge instead of re-deriving every query like RAG
- Public knowledge base that is multi-agent and multi-human auditable
- Provenance tracking with visible staleness decay for outdated content
- 25 MCP tools for programmatic wiki read/write access by external agents
- Self-healing pipeline with 6 specialized agents running on schedule
- Zero human code required—agent drives all development decisions
- 1,986 tests and 32 API routes all written by AI agent
- Markdown files readable in any markdown viewer without special tools
- Wikilinks between concepts create interlinked knowledge graph
- Confidence and expiry metadata on every page for trust assessment
- Open-source with Docker deployment and local development support
- Supports multiple LLM providers: Anthropic, OpenAI, Google, Ollama
- GitHub Issues serve as shared communication bus between agents
- Automatic revert on build failures with issue re-queuing
- Collision reconciliation on talk pages for contradictory content
- CLI and dark mode support for better developer experience
Yopedia cons
- No traditional pricing—requires self-hosting with your own LLM API costs
- Complex setup requiring TypeScript, Node.js, pnpm, and LLM API keys
- Only 57 stars on GitHub—very early stage with limited community
- Six-agent pipeline may be overkill for simple knowledge base needs
- Security relies on author allowlist—public repo potentially exploitable
- Agent-driven development means less human control over roadmap
- MCP server requires manual configuration in Claude Desktop or Cursor
- No cloud-hosted version available—must self-host entirely
- Learning curve for understanding agent communication via GitHub Issues
- Experimental project still evolving its agent surface design
Frequently asked questions about Yopedia
What is Yopedia?
Yopedia is a wiki designed for both humans and agents to read and write. It's a shared second brain for humans and agents with one knowledge substrate and two surfaces. The human surface uses Markdown files with YAML frontmatter, wikilinks, inline citations, and confidence/expiry metadata. The agent surface provides an MCP server with 25 tools for programmatic access. It grew from Karpathy's LLM Wiki pattern and was built entirely by an AI agent called yoyo with zero human code.
How is Yopedia different from RAG?
RAG re-derives every query from chunks each time, while Yopedia accumulates knowledge—pages update over time, contradictions reconcile on talk pages, lineage is preserved, and stale content visibly decays. Yopedia builds a persistent, interlinked wiki from raw sources rather than re-computing answers on every query.
How do the agents work?
Six specialized agents form a self-healing pipeline: Research (scans field for competitor intel, files max 3 issues), PM (reads vision, assesses gaps, files max 3 issues daily at 6am), Office Hour (triages issues—simple becomes [ready], complex becomes [needs-architecture]), Build (implements [ready] issues, opens PRs), Review (reviews diffs against acceptance criteria on PR open), and Architect (decomposes hard problems, diagnoses build failures). They communicate through GitHub Issues as a shared bus, running on GitHub Actions schedules.
What MCP tools are available?
Yopedia exposes 25 MCP tools across 8 categories: Wiki CRUD (search_wiki, read_page, list_pages, create_page, update_page, delete_page), Ingest (ingest_url, ingest_text, reingest), Query (query_wiki, save_query_answer, dataview_query), Lint (lint_wiki, fix_lint_issue), Discussions (list_discussions, create_discussion, resolve_discussion, add_comment), Agents (agent_context, seed_agent, list_agents, update_agent, delete_agent), and Revisions (list_revisions, read_revision). Configure in Claude Desktop/Cursor via mcpServers in config.
How do I run Yopedia locally?
Clone the repo with git clone https://github.com/yologdev/yopedia.git, cd into it, run pnpm install, create .env.local with your LLM API key (ANTHROPIC_API_KEY, OPENAI_API_KEY, GOOGLE_GENERATIVE_AI_KEY, or OLLAMA_BASE_URL for local), then run pnpm dev to start at http://localhost:3000. Supported providers are Anthropic (default: claude-sonnet-4-20250514), OpenAI (default: gpt-4o), Google Gemini (default: gemini-2.0-flash), and Ollama (default: llama3.2).
Is Yopedia secure?
Yes, with protections including random boundary nonces around issue content (unpredictable, unspoofable), content sanitization (HTML comments stripped, markers replaced), author allowlist (only approved users' issues get processed), protected files enforced mechanically after every task, and automatic revert if anything goes wrong. However, it's a public repo so anyone could file malicious issues—the harness is designed to handle this.
What LLM providers does Yopedia support?
Yopedia supports Anthropic (Claude, default model claude-sonnet-4-20250514), OpenAI (GPT, default model gpt-4o), Google (Gemini, default model gemini-2.0-flash), and Ollama (local, default model llama3.2). The app auto-detects provider from environment variables with priority: Anthropic → OpenAI → Google → Ollama. Set LLM_MODEL to override the default model name for your selected provider.
What is the project structure?
The repo contains llm-wiki.md (founding prompt, immutable), yopedia-concept.md (north star, immutable), SCHEMA.md (wiki conventions, LLM-readable), YOYO.md (project context + phased roadmap), .github/workflows/ with pm.yml, office-hour.yml, build.yml, review.yml, research.yml, architect.yml for agent schedules, src/ (everything written by agents), and .yoyo/ with yoyo.toml (agent config), skills/ (project-local agent skills), journal.md (what happened each session), and learnings.md (what agents learned about this project).