Obsidian Worktable

A native Obsidian dashboard plugin for focus, tasks, learning, review, and news. Built as a single custom ItemView (no Dataview dependency), six widgets render in a grid layout and share state through native IndexedDB. AI features can run either through the bundled local Cloakfetch service or via direct browser calls to Anthropic / OpenAI / Gemini-compatible endpoints.

🇨🇳 简体中文 → README_zh.md

---

Features

• Native ItemView dashboard — single view, no Dataview dependency, opens automatically on startup.
• Pomodoro focus timer — circular ring, custom durations, session history, daily focus averages (excluding today).
• Todo board — add, complete, and clear todos with priority levels; data persisted in IndexedDB.
• Inquiry-based learning — look up any concept; the AI expands it with translation, part of speech, and structured Markdown (great for vocabulary, math formulas, science notes).
• Active recall — AI-generates MCQ / cloze / true-false questions from an article and tracks per-knowledge-point mastery in plans/知识点.md.
• End-of-day review — surfaces items you wrote, archived, or marked for review; cross-widget "today" feed.
• News aggregation — surfaces any markdown inside the news folder or any file tagged #news anywhere in the vault, sorted by mtime.
• "小红花" encouragement widget — completion rewards and an archive of every archived learning record.
• Direct AI providers — Anthropic (Claude), OpenAI (GPT), Google Gemini, DeepSeek, Moonshot (Kimi), Zhipu (GLM), Alibaba Bailian (Qwen), Volcengine (Doubao), MiniMax — switchable in settings without code changes.
• Local Cloakfetch service for article fetching (/fetch) and as the AI proxy fallback; token auto-discovered from ~/.config/obsidian-worktable/server.json.
• Plain CSS — src/styles/*.css files are concatenated by scripts/build-styles.mjs; theme-friendly, no PostCSS step.

Installation

Option A: BRAT (recommended — works without submitting to Community Plugins)

1. Install BRAT from Obsidian's Community Plugins.
2. Open Settings → BRAT → Add a beta plugin.
3. Paste: https://github.com/DaoYoung/obsidian-worktable.git
4. Enable Worktable under Community Plugins.

BRAT installs from the latest published GitHub Release's attachments, not from the repo root. Releases must not be Draft, and must attach main.js, manifest.json, styles.css, and obsidian-worktable.zip.

If the repo is private, configure a GitHub token with read access in BRAT's SecretStorage and select that token when adding the plugin.

Option B: Manual install from GitHub Release

1. Download the latest obsidian-worktable.zip from Releases.
2. Unzip into your vault:

   vault/.obsidian/plugins/obsidian-worktable/
   

3. Enable Worktable under Community Plugins.

Option C: Build from source

git clone https://github.com/DaoYoung/obsidian-worktable.git
cd obsidian-worktable
npm install
npm run build

Copy main.js, manifest.json, and styles.css from the repo root into your vault's plugin directory. npm run dev watches for changes during development.

Once submitted and accepted into Obsidian's Community Plugins, this plugin will appear in Settings → Community plugins → Browse and can be installed with one click. Until then, BRAT is the supported install channel.

Local service (Cloakfetch)

Article fetching (/fetch) and the AI proxy when Direct AI is not configured both go through a small Python service bundled in server/server.py. Token auto-discovery means you usually do not have to copy anything into the plugin's settings — the installer writes the token to ~/.config/obsidian-worktable/server.json and the plugin reads it on first request.

macOS — one-shot install

Settings → Community plugins → Worktable → Options shows a "Setup local service" card. Click the install command, paste it into Terminal:

git clone https://github.com/DaoYoung/obsidian-worktable.git ~/obsidian-worktable
bash ~/obsidian-worktable/server/install-macos.sh

The script:

• Installs the Python service as a macOS launchd user agent (auto-start on login, restart on crash).
• Generates a random serviceToken and writes it to ~/.config/obsidian-worktable/server.json.
• Launches the service immediately.

After install, click Test connection in the same settings card to verify. A green "✓ Local service reachable" line means you're done.

Linux / Windows

launchd is macOS-only. On Linux and Windows, run the service manually:

git clone https://github.com/DaoYoung/obsidian-worktable.git ~/obsidian-worktable
cd ~/obsidian-worktable/server
pip3 install -r requirements.txt   # first time only
python3 server.py

Leave the terminal open, or run under tmux / systemd-user / NSSM. serviceToken auth works the same way; just put a value into ~/.config/obsidian-worktable/server.json (or set WORKTABLE_SERVICE_TOKEN) and the plugin picks it up.

Configuration file

~/.config/obsidian-worktable/server.json:

{
  "host": "127.0.0.1",
  "port": 8765,
  "serviceToken": "your-service-token-here",
  "aiAuthToken": "sk-ant-api03-xxxxxxxx",
  "aiBaseUrl": "https://api.anthropic.com",
  "aiModel": "claude-sonnet-4-5",
  "aiMaxTokens": 2048,
  "aiTimeout": 60,
  "upstreamFetchTimeout": 30
}

• host — listen address. Do not change to 0.0.0.0; the plugin talks to the service over loopback only.
• port — defaults to 8765. Change if you have a port conflict.
• serviceToken — required header X-Worktable-Token for every request. Leave empty to disable auth (fine for single-user, no-trust local).
• aiAuthToken — Anthropic-compatible key (also accepts anthropicApiKey). Env vars ANTHROPIC_AUTH_TOKEN / ANTHROPIC_API_KEY / ~/.claude/settings.json override the file.
• aiBaseUrl / aiModel — defaults follow the model's defaults but can be repointed to any compatible gateway.

Any field can be overridden with WORKTABLE_<FIELD_NAME> env vars (camelCase → UPPER_SNAKE), which take precedence over the file:

export WORKTABLE_SERVICE_TOKEN="your-token"
export WORKTABLE_AI_BASE_URL="https://your-proxy.example.com"
export WORKTABLE_PORT=9001

Token auto-discovery

The plugin reads the service token in this order:

1. The "Service token (advanced override)" field in plugin settings — leave empty unless bypassing the config file.
2. ~/.config/obsidian-worktable/server.json serviceToken.
3. /etc/obsidian-worktable/server.json serviceToken.
4. Empty string → server's _check_auth allows the request.

After running install-macos.sh, step 2 is set automatically and you do not need to touch plugin settings.

Plugin settings

Settings → Community plugins → Obsidian Worktable → Options:

Setting	Default	Notes
Knowledge file	plans/知识点.md	Vault-relative path the learning modules read and write
News folder	news	Folder for the news widget; #news-tagged files anywhere also appear
Service base URL	http://127.0.0.1:8765	Local Cloakfetch service endpoint
Service token (advanced override)	""	Leave empty — auto-discovered from ~/.config/obsidian-worktable/server.json
Open on startup	✅	Auto-open the Worktable view when Obsidian starts
Enable fallback proxies	✅	CORS proxy fallback when article fetching fails
Provider	anthropic	AI provider for Direct AI mode
API key	""	Provider API key
Base URL	https://api.anthropic.com	Auto-fills on provider change; supports private gateways
Model	claude-sonnet-4-5	Auto-fills on provider change; manual override allowed

Settings live in .obsidian/plugins/obsidian-worktable/data.json, managed by Obsidian.

Direct AI vs local service

When API key + Base URL + Model are all filled, the plugin calls your provider directly from the browser (works with native CORS-enabled endpoints: Anthropic, OpenAI, Google Gemini). /fetch still goes through the local Cloakfetch service.

If any of those three is empty, AI calls fall through to the local service's /ai/* endpoints (which read the same provider config from server.json server-side).

AI providers

Provider	Default Base URL	Default Model
Anthropic (Claude)	https://api.anthropic.com	claude-sonnet-4-5
OpenAI (GPT)	https://api.openai.com/v1	gpt-4o-mini
Google Gemini	https://generativelanguage.googleapis.com/v1beta	gemini-2.0-flash
DeepSeek	https://api.deepseek.com/v1	deepseek-chat
Moonshot (Kimi)	https://api.moonshot.cn/v1	moonshot-v1-8k
Zhipu (GLM)	https://open.bigmodel.cn/api/paas/v4	glm-4-flash
Alibaba Bailian (Qwen)	https://dashscope.aliyuncs.com/compatible-mode/v1	qwen-plus
Volcengine (Doubao)	https://ark.cn-beijing.volces.com/api/v3	doubao-1-5-pro-32k-250115
MiniMax	https://api.minimaxi.com/anthropic	MiniMax-M3

Any OpenAI Chat-Completions-compatible endpoint works: pick openai and override Base URL. Switching providers auto-fills Base URL and Model; you can override either afterwards.

Troubleshooting

Plugin fails to enable

1. Confirm Obsidian version ≥ 1.5.0 (Settings → About).
2. Check main.js, manifest.json, styles.css are all present.
3. Open Ctrl/Cmd + P → "Show debug console", reload, look for red errors mentioning worktable.

"Local service is not reachable: ECONNREFUSED"

Run the appropriate install step above, then Test connection in the plugin settings card. If the service is running but the plugin can't reach it, the Service base URL in plugin settings is probably wrong.

News widget shows nothing

The widget scans the configured folder plus any markdown with the #news tag. Either create news/ in the vault and add a file, or add #news to an existing note's frontmatter / inline.

Storage and reset

All widget data lives in IndexedDB under the database home-db and the localStorage keys documented in src/widgets/*.ts. Use the Obsidian debug console → "Inspect" to clear them, or uninstall + reinstall the plugin (Obsidian prompts before deleting storage).

Development

npm install
npm run dev          # watch mode (CSS + esbuild)
npm run typecheck    # tsc --noEmit
npm test             # vitest run
npm run test:server  # python3 -m unittest in server/tests
npm run verify       # typecheck + tests + build + artifact checks

Source layout:

• src/main.ts — plugin entry, view registration
• src/view/WorktableView.ts — single ItemView, grid layout, widget descriptor registry
• src/widgets/<name>Widget.ts and src/widgets/<name>.ts — each widget pair (logic + lazy-load wrapper)
• src/services/CloakfetchClient.ts — the only AI transport; routes Direct AI vs local service
• src/services/ai/{registry,anthropic,openaiCompat,gemini,client}.ts — provider specs
• src/styles/{base,productivity,learning}.css — concatenated by scripts/build-styles.mjs to root styles.css
• server/ — Python Cloakfetch service (FastAPI / uvicorn)
• tests/ — Vitest specs + server/tests/ for the Python side

Privacy

• All widget data is stored locally in your vault (IndexedDB + the file plans/知识点.md for knowledge points).
• The plugin does not collect telemetry. No analytics calls.
• Tokens and API keys live in .obsidian/plugins/obsidian-worktable/data.json (managed by Obsidian) or ~/.config/obsidian-worktable/server.json, and are sent only to the local service over loopback.
• server.json is in .gitignore — no secrets reach the repo.

License

MIT