🍴 NomFeed

Nom the web. Feed your agents.

NomFeed converts any URL, YouTube video, or file into clean, structured markdown — and keeps it in a searchable local library. One CLI command to save. One command to search. Optional LLM extraction turns content into structured intelligence.

nomfeed add https://example.com/article          # URL → markdown
nomfeed add https://youtube.com/watch?v=xyz       # YouTube transcript
nomfeed add ./report.pdf                          # File → markdown
nomfeed search "transformers"                     # Full-text search
nomfeed read AMxTn2tS5_ --extract                 # Read LLM extraction

No database. No Docker. No daemon. Just .md files in ~/.nomfeed/.

---

How It Compares

NomFeed sits in the same space as markdown.new and keep.md — tools that convert web content to markdown for AI consumption. We love what they've built. NomFeed takes a different approach:

	markdown.new	keep.md	NomFeed
What it does	URL → markdown (single conversion)	Bookmark + markdown API	CLI library + extraction engine
Storage	None (stateless)	Cloud	Local flat files (~/.nomfeed/)
YouTube	❌	❌	✅ Full transcript via yt-dlp
File conversion	✅ 20+ formats	❌	✅ PDF, DOCX, XLSX, code, images
LLM extraction	❌	❌	✅ 6 patterns (wisdom, claims, chapters...)
Search	❌	✅ API-based	✅ Local full-text search
MCP server	❌	✅	✅ 7 tools
Chrome extension	❌	✅	✅ Popup + context menu
Offline	❌	❌	✅ Everything local after save
Data ownership	N/A	Their servers	Your filesystem
Cost	Free	Free tier + paid	Free forever (BYO LLM key)
URL strategies	Cloudflare only	Unknown	3-strategy cascade

markdown.new is perfect when you need a quick one-off conversion. keep.md is great if you want a hosted API. NomFeed is for developers who want a local library they own, with search, extraction, and agent integration — all on their own machine.

---

Install

git clone https://github.com/ameno-/nomfeed
cd nomfeed
bun install
bun link

# Optional: file conversion (PDF, DOCX, XLSX, images)
pip install 'markitdown[all]'

# Optional: YouTube transcript extraction
brew install yt-dlp

# Optional: LLM extraction (--extract flag)
export OPENROUTER_API_KEY=your-key  # get one at openrouter.ai/keys

CLI

# Save content
nomfeed add <url>                    # URL → markdown
nomfeed add <url> --extract          # URL → markdown + LLM extraction
nomfeed add <file>                   # File → markdown
nomfeed note "some text"             # Quick note

# Retrieve
nomfeed list                         # List everything (✦ = has extraction)
nomfeed read <id>                    # Print markdown content
nomfeed read <id> --extract          # Print extraction only
nomfeed read <id> --full             # Print both
nomfeed search "query"               # Full-text search

# Extract
nomfeed extract <id>                 # Run extraction on existing item
nomfeed extract <id> --patterns summarize,analyze_claims
nomfeed patterns                     # List available patterns

# Manage
nomfeed delete <id>
nomfeed status

Every command supports --json for machine-readable output.

URL Conversion

NomFeed uses a 3-strategy cascade — the first success wins:

1. Cloudflare text/markdown — native edge conversion, fastest
2. Jina Reader — renders JavaScript SPAs, returns clean markdown
3. Readability — universal fallback, extracts article content from raw HTML

This means NomFeed works on sites that break simpler tools: JS-heavy SPAs, paywalled articles (where possible), and pages that return empty HTML without rendering.

LLM Extraction

When you add with --extract, NomFeed runs Fabric-inspired patterns against the content via LLM. Each pattern is a single focused API call — no agents, no orchestration, just reliable structured output.

Pattern	What It Extracts
extract_wisdom	Ideas, insights, quotes, habits, facts, references, recommendations
video_chapters	Timestamped chapter outline
analyze_claims	Truth claims with evidence ratings (A–F) and logical fallacies
extract_references	Books, papers, tools, people mentioned
summarize	One-sentence summary + main points + takeaways
rate_content	Quality scores: surprise, novelty, insight, value, wisdom (0–10)

# Default patterns: extract_wisdom + video_chapters
nomfeed add https://youtube.com/watch?v=xyz --extract

# Choose specific patterns
nomfeed extract <id> --patterns extract_wisdom,analyze_claims,rate_content

# Add your own
# Create ~/.nomfeed/patterns/<name>/system.md with your prompt
nomfeed patterns  # will show custom patterns

Provider: OpenRouter — one API key, access to all models. Default model chain: Claude Sonnet 4.5 → Sonnet 4 → Haiku 4.5 (automatic fallback on rate limits/errors).

MCP Server

nomfeed mcp  # starts stdio MCP server

Tools: nomfeed_add, nomfeed_list, nomfeed_read (content/extract/full), nomfeed_search, nomfeed_extract, nomfeed_delete, nomfeed_status

Claude Desktop / Cursor / Windsurf Config

{
  "mcpServers": {
    "nomfeed": {
      "command": "bun",
      "args": ["run", "/path/to/nomfeed/src/cli.ts", "mcp"]
    }
  }
}

Chrome Extension

1. Run nomfeed serve (starts local HTTP server on port 24242)
2. chrome://extensions → Developer Mode → Load unpacked → select extension/
3. Click the NomFeed icon or right-click any page → "Save to NomFeed"

The extension supports tagging and optional extraction on save.

HTTP API

nomfeed serve          # default port 24242
nomfeed serve --port 8080

Method	Path	Description
POST	/add	Save URL, file, or note. { url, title?, tags?, extract? }
GET	/items	List items. ?query=&tag=&type=&limit=
GET	/items/:id	Get item metadata + content
DELETE	/items/:id	Delete item
GET	/search?q=	Full-text search
GET	/health	Health check

Storage

Everything lives in ~/.nomfeed/ (override with NOMFEED_DIR):

~/.nomfeed/
├── items.json              # metadata index (array of items)
└── content/
    ├── abc123.md           # converted markdown
    ├── abc123.extraction.md # LLM extraction (if extracted)
    └── def456.md

Each .md file includes YAML frontmatter with source, title, and timestamp. Files are self-contained — you can copy them anywhere and they still make sense.

Environment Variables

Variable	Description	Required
OPENROUTER_API_KEY	LLM access for --extract	Only for extraction
NOMFEED_MODEL	Override default model	No
NOMFEED_DIR	Override data directory	No (default: ~/.nomfeed)

Agent Skill

NomFeed ships with a Pi / Agent Skills compatible skill in skills/nomfeed/. Install it so your coding agent can save, search, and extract content on your behalf.

Install the skill

# Copy to your global skills directory
cp -r skills/nomfeed ~/.pi/agent/skills/nomfeed

# Or symlink it (stays in sync with repo updates)
ln -sf "$(pwd)/skills/nomfeed" ~/.pi/agent/skills/nomfeed

The skill is also compatible with other harnesses that support the Agent Skills standard:

# Claude Code
cp -r skills/nomfeed ~/.claude/skills/nomfeed

# Codex
cp -r skills/nomfeed ~/.codex/skills/nomfeed

Once installed, your agent will automatically trigger NomFeed on phrases like "save this URL", "bookmark this", "search my saved content", "extract insights from this video", etc.

Architecture

See DESIGN.md for the full system architecture, conversion strategies, extraction pipeline, and design decisions.

License

MIT