Your AI Research Platform — From Papers to Publications, Powered by Claude Code
From paper ingestion to publication — your research knowledge compounds, never decays.
ΩmegaWiki is built by DAIR Lab at Peking University — a fully agentic platform that automates the complete research pipeline, from knowledge ingestion to paper submission.
Weitong Qian PKU Undergraduate · 2023 |
Beicheng Xu PKU Ph.D. · 2023 |
Zhongao Xie PKU Undergraduate · 2025 |
Bowen Fan PKU Undergraduate · 2024 |
Guozheng Tang PKU Undergraduate · 2024 |
Xinzhe Wu PKU Undergraduate · 2024 |
Jiale Chen PKU Undergraduate · 2024 |
Mingtian Yang PKU Undergraduate · 2024 |
Run /discover --venue iclr --year 2024 (or any conference/year) and get a personalized shortlist of papers from that venue, ranked by relevance to what's already in your wiki. Instead of scrolling a 7000-paper proceedings, you see the dozen that actually matter for your research direction, each with a rationale tied to topics and methods you already track. No new API keys, no ingest side-effects on your wiki — just a ranked reading list. Supports NeurIPS, ICLR, ICML, and other venues covered by Paper Copilot.
Run /daily-arxiv for a one-off pass, or /daily-arxiv setup to schedule the same pipeline in GitHub Actions. The skill builds an evidence packet from arXiv + Semantic Scholar + DeepXiv, lets the LLM rank candidates against your wiki interests, and delivers a digest by e-mail. Explicit --mode auto-ingest calls /ingest for high-confidence picks; inform mode just notifies.
Your research graph now has two ways to explore:
- Web UI — run
python3 tools/serve.py, openhttp://localhost:8765/#/graph. Click any node to highlight its neighborhood via BFS, filter by entity type or edge category, double-click to open the full page in the Reader. - Obsidian — run
/visualize --obsidianto generate a color-coded graph config, or/visualize --canvasto produce a force-layout Canvas with labeled semantic edges.
Architectures, training recipes, evaluation protocols, and other reusable techniques now live in wiki/methods/ as proper wiki entities — with their own pages, source-paper links, and parent/child method chains.
Andrej Karpathy proposed LLM-Wiki: an LLM that builds and maintains a persistent, structured wiki from your sources — not a throwaway RAG answer, but compounding knowledge that grows smarter with every paper you feed it.
ΩmegaWiki takes that idea and runs the full distance. It's not just a wiki builder — it's a complete research lifecycle platform: from paper ingestion → knowledge graph → gap detection → idea generation → experiment design → paper writing → peer review response. All driven by 24 Claude Code skills, all centered on one wiki as the single source of truth.
Drop your .tex / .pdf files in a folder. Run one command. Get a fully cross-referenced knowledge base — and then use it to generate novel research ideas, design experiments, write papers, and respond to reviewers.
| RAG | ΩmegaWiki | |
|---|---|---|
| Knowledge persistence | Rediscovered on every query | Compiled once, maintained forever |
| Structure | Flat chunk store | 9 typed entities with relationships |
| Cross-references | None — chunks are isolated | Bidirectional wikilinks + typed graph |
| Knowledge gaps | Invisible | Explicitly tracked, drive research |
| Failed experiments | Lost | First-class anti-repetition memory |
| Output | Chat answers | Papers, surveys, experiment plans, rebuttals |
| Compounding | No — same cost every query | Yes — each paper enriches the whole graph |
Every skill reads from and writes back to the wiki. Knowledge compounds — each new paper enriches the whole graph. Failed experiments aren't discarded; they become anti-repetition memory that prevents re-exploring dead ends.
Prerequisites: Python 3.9+, Node.js 18+
# 1. Clone
git clone https://github.com/skyllwt/OmegaWiki.git
cd OmegaWiki
# 2. Install Claude Code
npm install -g @anthropic-ai/claude-code
claude login
# 3. One-click setup
chmod +x setup.sh && ./setup.sh # Linux / macOS
# Windows (PowerShell):
# powershell -ExecutionPolicy Bypass -File .\setup.ps1
# setup creates .venv for OmegaWiki
# the script does not keep your shell activated, but /init will use .venv automatically
# 4. Put your own papers in raw/papers/ (.tex or .pdf)
# Optional: add intent notes to raw/notes/ and saved pages to raw/web/
# /init and direct local /ingest will manage generated inputs under raw/discovered/ and raw/tmp/
# 5. Build your wiki
claude
# Then type: /init [your-research-topic]Manual setup (Linux / macOS)
python3 -m venv .venv && source .venv/bin/activate
pip install -r requirements.txt
cp .env.example .env # Edit to add API keys
cp config/settings.local.json.example .claude/settings.local.jsonManual setup (Windows / PowerShell)
python -m venv .venv
.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt
Copy-Item .env.example .env # Edit to add API keys
Copy-Item config\settings.local.json.example .claude\settings.local.jsonNote: native Windows is supported for the local pipeline. Remote-GPU
experiments via /exp-run --env remote rely on ssh/rsync/screen
and are best run from WSL2 or Linux/macOS.
| Key | Required? | How to get | What it enables |
|---|---|---|---|
ANTHROPIC_API_KEY |
Yes | claude login (automatic) |
Powers all Claude Code skills |
CLAUDE_CODE_OAUTH_TOKEN |
Optional | claude setup-token |
GitHub Actions Claude Code auth for Pro/Max users |
SEMANTIC_SCHOLAR_API_KEY |
Optional | semanticscholar.org/product/api (free) | Citation graph, paper search |
DEEPXIV_TOKEN |
Optional | setup.sh auto-registers |
Semantic search, TLDR, trending |
LLM_API_KEY + LLM_BASE_URL + LLM_MODEL |
Optional | Any OpenAI-compatible API | Cross-model review; /daily-arxiv inform recommendations |
Cross-model review: ΩmegaWiki uses a second LLM as an independent reviewer for ideas, experiments, and paper drafts. Works with any OpenAI-compatible API — DeepSeek, OpenAI, Qwen, OpenRouter, SiliconFlow, etc. If not configured, skills still work in Claude-only mode.
/daily-arxiv runs a one-off fresh-paper recommendation pass even before
automation is configured. To schedule the same pipeline in GitHub Actions, copy
config/daily-arxiv.yml.example to config/daily-arxiv.yml, then run
/daily-arxiv setup. The config stores non-secret preferences such as mode,
categories, caps, and schedule; SMTP/API credentials stay in .env or GitHub
Actions secrets. In CI inform mode, recommendations can use Claude Code auth
(ANTHROPIC_API_KEY or CLAUDE_CODE_OAUTH_TOKEN) or the OpenAI-compatible
LLM_* review model; auto-ingest still requires Claude Code.
See
docs/daily-arxiv-deployment.mdfor the GitHub Actions setup checklist and symptom-keyed troubleshooting.
Sample digest
A real /daily-arxiv run: ranked recommendations with scores, rationales, wiki connections, and an auto-ingest section.
24 slash commands spanning the full research lifecycle:
| Command | What it does |
|---|---|
/setup |
First-time configuration (API keys, language, dependencies) |
/reset <scope> |
Destructive cleanup: wiki | raw | log | checkpoints | all |
| Command | What it does |
|---|---|
/prefill <domain> |
Optionally seed foundations/ with background knowledge |
/init [topic] |
Bootstrap a full wiki from user raw sources plus optional discovery |
/ingest <source> |
Parse a paper → wiki pages + cross-references |
/discover |
Recommend ranked next-read papers from anchors, a topic, the current wiki, or a venue/year |
/edit <request> |
Add/remove sources or update wiki content |
/ask <question> |
Query the wiki, crystallize answers back |
/check |
Health scan: broken links, missing cross-refs, consistency |
| Command | What it does |
|---|---|
/daily-arxiv |
Run/manage a daily arXiv recommendation feed (+ optional GitHub Actions scheduler) |
/ideate |
Multi-phase idea generation from cross-topic connections |
/novelty <idea> |
Multi-source novelty verification (web + S2 + wiki + review LLM) |
/review <artifact> |
Cross-model adversarial review for any research artifact |
/exp-design <idea> |
Idea-driven experiment + ablation design |
/exp-run <experiment> |
Implement + deploy + monitor (local or remote GPU) |
/exp-status |
Dashboard for running experiments; auto-collect results |
/exp-eval <experiment> |
Verdict gate → auto-update the linked idea + graph |
/refine <artifact> |
Multi-round: produce → review → fix → re-review |
| Command | What it does |
|---|---|
/survey |
Generate Related Work from wiki knowledge |
/paper-plan <ideas> |
Outline from validated-idea graph + evidence matrix |
/paper-draft <plan> |
Draft LaTeX + figures, section by section |
/paper-compile <dir> |
Compile → PDF, auto-fix, verify page/anonymity |
/research <direction> |
End-to-end orchestrator with human gates |
/rebuttal <reviews> |
Parse reviewer comments → draft point-by-point responses |
| Type | Directory | Purpose |
|---|---|---|
| Paper | papers/ |
Structured summary: problem/key idea/method/experiment+results/limitations + tldr/contribution_type/datasets |
| Concept | concepts/ |
Cross-paper technical concept with variants, comparisons, definition, linked ideas |
| Topic | topics/ |
Research direction map with SOTA tracker, key benchmarks, and open problems (split into known + methodological gaps) |
| Person | people/ |
Researcher profile with research areas, recent work, and a researcher/team/organization type |
| Idea | ideas/ |
Research idea with lifecycle, novelty argument & score, target venue |
| Experiment | experiments/ |
Full record: hypothesis → setup → results → updates to the linked idea |
| Method | methods/ |
Reusable, citable technique entity (cross-paper); links to source papers and parent/child methods |
| Summary | Summary/ |
Domain-wide survey across topics |
| Foundation | foundations/ |
Background knowledge (terminal: receives inward links, writes none) |
Semantic relationships are stored in graph/edges.jsonl; bibliographic paper citations are stored separately in graph/citations.jsonl.
Paper-paper semantic edges include same_problem_as, similar_method_to, complementary_to, builds_on, compares_against, improves_on, challenges, and surveys. Paper-concept edges use introduces_concept, uses_concept, extends_concept, and critiques_concept. Workflow edges (supports, contradicts, tested_by, invalidates, addresses_gap, inspired_by, derived_from) span experiments, ideas, methods, and concepts.
All pages use Obsidian [[wikilink]] format — open wiki/ in Obsidian for visual graph exploration.
GitHub Actions runs the /daily-arxiv recommendation pipeline at UTC 00:17 daily (08:17 Beijing time):
- Add SMTP secrets to repo Settings → Secrets when e-mail delivery is enabled:
SMTP_HOST,SMTP_PORT,SMTP_USER,SMTP_PASSWORD,SMTP_FROM,DAILY_ARXIV_EMAIL_TO - Optional inform-mode LLM recommendation: add
ANTHROPIC_API_KEYorCLAUDE_CODE_OAUTH_TOKENfor Claude Code, orLLM_API_KEY,LLM_BASE_URL, andLLM_MODELfor any OpenAI-compatible provider .github/workflows/daily-arxiv.ymlfetches arXiv, deduplicates against the wiki, builds a recommendation context, uploads artifacts, and sends the digest by SMTP
auto-ingest mode is explicit and requires Claude Code in CI, because plain API LLMs cannot invoke slash skills such as /ingest. Use manual dispatch with send_email=false for a dry run without SMTP secrets.
OmegaWiki/
├── CLAUDE.md # Runtime schema & rules
├── wiki/ # Knowledge base (LLM-maintained)
│ ├── papers/ # Structured paper summaries
│ ├── concepts/ # Cross-paper technical concepts
│ ├── topics/ # Research direction maps
│ ├── people/ # Researcher profiles
│ ├── ideas/ # Research ideas (with lifecycle)
│ ├── experiments/ # Experiment records
│ ├── methods/ # Reusable cross-paper method entities
│ ├── Summary/ # Domain-wide surveys
│ ├── foundations/ # Background knowledge (terminal pages)
│ ├── outputs/ # Generated artifacts
│ ├── graph/ # Auto-generated: edges, context, gaps
│ ├── index.md # Content catalog
│ └── log.md # Chronological log
├── raw/ # Source materials
│ ├── papers/ # User-owned .tex / .pdf files
│ ├── discovered/ # external papers from /init and explicit /daily-arxiv auto-ingest
│ ├── tmp/ # generated prepared local sidecars for /init and direct local /ingest
│ ├── notes/ # User-owned .md notes
│ └── web/ # User-owned HTML / Markdown
├── tools/ # Deterministic Python helpers
│ ├── research_wiki.py # Wiki engine (20 CLI commands)
│ ├── init_discovery.py # /init prepare + plan + fetch helper
│ ├── discover.py # /discover candidate gathering, dedup, ranking
│ ├── lint.py # Structural validation (10 checks)
│ ├── reset_wiki.py # Scoped destructive cleanup helper
│ ├── fetch_arxiv.py # arXiv RSS fetcher
│ ├── fetch_s2.py # Semantic Scholar API
│ ├── fetch_deepxiv.py # DeepXiv semantic search
│ ├── fetch_wikipedia.py # Wikipedia fetcher (used by /prefill)
│ └── remote.py # SSH ops for remote experiments
├── .claude/skills/ # 24 Claude Code skill definitions
├── i18n/ # Bilingual: en/ (canonical) + zh/
├── config/ # Configuration templates
├── mcp-servers/ # Cross-model review server
└── .github/workflows/ # Daily arXiv cron
ΩmegaWiki ships in English and Chinese:
./setup.sh --lang en # English (default)
./setup.sh --lang zh # 中文We welcome contributions! See CONTRIBUTING.md for guidelines.
ΩmegaWiki runs on Claude Code, which speaks the Anthropic API protocol. You can use Claude directly, or route Claude Code to any third-party provider that exposes an Anthropic-compatible endpoint by overriding a few environment variables.
ΩmegaWiki 基于 Claude Code,Claude Code 使用 Anthropic API 协议通信。你既可以直接使用 Claude,也可以通过覆盖几个环境变量,把 Claude Code 指向任意支持 Anthropic 协议的第三方供应商。
claude login # OAuth, no manual config / OAuth 登录,无需手动配置Pick a provider below, paste the snippet into ~/.claude/settings.json (or the project's .claude/settings.json), and replace the <...> placeholder with your own API key. Model names and extra options are taken from each provider's official Claude Code docs — if anything stops working (e.g. a model is renamed), check the provider's website.
从下方任选一个供应商,把对应配置粘贴到 ~/.claude/settings.json(或项目的 .claude/settings.json),并把 <...> 占位符替换为你自己的 API key。模型名与额外选项均来自各供应商官方 Claude Code 文档;若出现问题(例如模型改名),请查询对应官网。
{
"env": {
"ANTHROPIC_BASE_URL": "https://api.xiaomimimo.com/anthropic",
"ANTHROPIC_AUTH_TOKEN": "<your-mimo-key>",
"ANTHROPIC_MODEL": "mimo-v2.5",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "mimo-v2.5",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "mimo-v2.5-pro",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "mimo-v2.5"
}
}{
"env": {
"ANTHROPIC_BASE_URL": "https://api.deepseek.com/anthropic",
"ANTHROPIC_AUTH_TOKEN": "<your-deepseek-key>",
"ANTHROPIC_MODEL": "deepseek-v4-pro[1m]",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "deepseek-v4-pro[1m]",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "deepseek-v4-pro[1m]",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "deepseek-v4-flash",
"CLAUDE_CODE_SUBAGENT_MODEL": "deepseek-v4-flash",
"CLAUDE_CODE_EFFORT_LEVEL": "max"
}
}{
"env": {
"ANTHROPIC_BASE_URL": "https://api.moonshot.ai/anthropic",
"ANTHROPIC_AUTH_TOKEN": "<your-moonshot-key>",
"ANTHROPIC_MODEL": "kimi-k2.5",
"ANTHROPIC_DEFAULT_OPUS_MODEL": "kimi-k2.5",
"ANTHROPIC_DEFAULT_SONNET_MODEL": "kimi-k2.5",
"ANTHROPIC_DEFAULT_HAIKU_MODEL": "kimi-k2.5",
"CLAUDE_CODE_SUBAGENT_MODEL": "kimi-k2.5",
"ENABLE_TOOL_SEARCH": "false"
}
}{
"env": {
"ANTHROPIC_BASE_URL": "https://api.z.ai/api/anthropic",
"ANTHROPIC_AUTH_TOKEN": "<your-zai-key>",
"API_TIMEOUT_MS": "3000000"
}
}Z.AI applies a default server-side model mapping, so no explicit
ANTHROPIC_MODELis needed. Z.AI 默认在服务端做模型映射,无需显式设置ANTHROPIC_MODEL。
Skip the Claude Code onboarding / 跳过 Claude Code 初始引导
When using a third-party key (instead of claude login), Claude Code's first-run onboarding won't complete automatically. Create or edit .claude.json and mark it done:
使用第三方 key 时不会走 claude login,Claude Code 首次启动的引导不会自动完成。创建或编辑 .claude.json,手动标记引导已完成:
- macOS / Linux:
~/.claude.json - Windows:
<user-home>\.claude.json
{
"hasCompletedOnboarding": true
}Then run claude as usual. / 保存后正常运行 claude 即可。
Scan to join the ΩmegaWiki WeChat group / 扫码加入微信交流群
- Andrej Karpathy — for the LLM-Wiki concept that inspired this project
- Claude Code — the AI agent runtime that powers ΩmegaWiki
MIT — use it, fork it, build on it.
Andrej Karpathy 提出了 LLM-Wiki 概念：让 LLM 构建并维护一个持久的、结构化的 wiki，而不是一次性的 RAG 回答。知识持续积累，每一篇新论文都让整个知识图谱更强。
ΩmegaWiki 将这个理念完整实现。 它不仅是 wiki 构建器，更是完整的研究全流程平台：从论文摄入 → 知识图谱 → 缺口检测 → 想法生成 → 实验设计 → 论文写作 → 同行评审回复。24 个 Claude Code Skills 驱动，一个 wiki 作为唯一的知识中枢。
| RAG | ΩmegaWiki | |
|---|---|---|
| 知识持久性 | 每次查询都重新发现 | 编译一次，持续维护 |
| 结构 | 扁平的 chunk 存储 | 9 种实体类型 + 关系图 |
| 交叉引用 | 无 — chunk 彼此孤立 | 双向 wikilink + 类型化边 |
| 知识缺口 | 不可见 | 显式追踪，驱动研究方向 |
| 失败实验 | 丢失 | 一等公民，防止重复探索 |
| 输出 | 聊天回答 | 论文、综述、实验方案、审稿回复 |
| 复利效应 | 无 — 每次查询成本相同 | 有 — 每篇论文丰富整个图谱 |
前置条件： Python 3.9+, Node.js 18+
git clone https://github.com/skyllwt/OmegaWiki.git && cd OmegaWiki
# 安装 Claude Code
npm install -g @anthropic-ai/claude-code
claude login
# 一键配置
chmod +x setup.sh && ./setup.sh --lang zh # Linux / macOS
# Windows (PowerShell):
# powershell -ExecutionPolicy Bypass -File .\setup.ps1 -Lang zh
# setup 会为 OmegaWiki 创建 .venv
# 脚本不会把你当前 shell 永久激活，但 /init 会自动使用 .venv
# 把你自己的论文放入 raw/papers/（.tex 或 .pdf）
# 可选：把意图笔记放入 raw/notes/，网页存档放入 raw/web/
# /init 与直接本地 /ingest 会自动管理 raw/discovered/ 与 raw/tmp/ 下的生成内容
# 启动 Claude Code
claude
# 输入：/init [你的研究方向]Windows 用户：本地 pipeline 已原生支持。
/exp-run --env remote远程 GPU 实验依赖ssh/rsync/screen，建议在 WSL2 或 Linux/macOS 下运行。
| Key | 必须？ | 获取方式 | 用途 |
|---|---|---|---|
ANTHROPIC_API_KEY |
是 | claude login |
驱动所有 Skill |
CLAUDE_CODE_OAUTH_TOKEN |
可选 | claude setup-token |
Pro/Max 用户的 GitHub Actions Claude Code auth |
SEMANTIC_SCHOLAR_API_KEY |
可选 | semanticscholar.org（免费） | 引用图谱、论文搜索 |
DEEPXIV_TOKEN |
可选 | setup.sh 自动注册 |
语义搜索、热门趋势 |
LLM_API_KEY + LLM_BASE_URL + LLM_MODEL |
可选 | 任意 OpenAI 兼容 API | 跨模型评审；/daily-arxiv inform 推荐 |
GitHub Actions 每天 UTC 00:17（北京时间 08:17）运行 /daily-arxiv 推荐 pipeline：拉取 arXiv、按 wiki 去重、构建 recommendation context、上传 artifacts，并可通过 SMTP 发送 digest 邮件。
启用邮件时，在 repo Settings → Secrets 添加：SMTP_HOST、SMTP_PORT、SMTP_USER、SMTP_PASSWORD、SMTP_FROM、DAILY_ARXIV_EMAIL_TO。
CI inform mode 可使用 ANTHROPIC_API_KEY 或 CLAUDE_CODE_OAUTH_TOKEN 启动 Claude Code，也可使用 LLM_API_KEY、LLM_BASE_URL、LLM_MODEL 接入任意 OpenAI-compatible provider。auto-ingest 是显式模式，并且需要 Claude Code，因为普通 API LLM 不能调用 /ingest 这类 slash skill。手动触发时可设置 send_email=false，用于无 SMTP secrets 的 dry run。
Digest 示例 / Sample digest
一次真实的 /daily-arxiv 运行结果：带分数、理由、wiki 关联以及 auto-ingest 区块的推荐 digest。
| 命令 | 功能 |
|---|---|
/setup |
首次配置（API key、语言、依赖） |
/reset |
按范围销毁性清理：wiki | raw | log | checkpoints | all |
/prefill |
可选地预填 foundations/ 背景知识 |
/init |
基于用户 raw 素材并按需做外部发现来搭建 wiki |
/ingest |
消化论文，创建页面 + 交叉引用 |
/discover |
从 anchor、topic、当前 wiki 或 venue/year 推荐排序后的下一批待读论文 |
/edit |
增删 raw 或更新 wiki |
/ask |
对 wiki 提问 |
/check |
wiki 健康检查 |
/daily-arxiv |
运行/管理每日 arXiv 推荐 feed（可选 CI 定时） |
/ideate |
跨方向构思研究 idea |
/novelty |
多源新颖性验证 |
/review |
跨模型评审 |
/exp-design |
idea 驱动的实验设计 |
/exp-run |
部署 + 监控实验 |
/exp-status |
实验状态看板 |
/exp-eval |
裁决 → 自动更新关联 idea |
/refine |
多轮迭代改进 |
/survey |
生成 Related Work |
/paper-plan |
idea 图谱 + 实验证据 → 论文提纲 |
/paper-draft |
提纲 + wiki → LaTeX 草稿 |
/paper-compile |
编译 → PDF，自动修复 |
/research |
端到端研究编排器 |
/rebuttal |
解析评审意见 → 逐条回复 |
Built with Claude Code
If this project helps your research, give it a ⭐