raycast-vault-clipper

Two Raycast scripts for sending links from your active browser tab into an Obsidian vault.

Command	Script	Use it for
Save to Vault	save-to-vault.py	Comprehensive saved notes with extracted content, AI summary, tags, and frontmatter
Read Later to Daily	read-later.py	One lightweight bookmark link appended to today's daily note

Use Save to Vault when you want a full note you can search, process, and revisit. Use Read Later to Daily when you only want to park a link for later.

---

Why this exists

The official Obsidian Web Clipper is a browser extension, which can be blocked in locked-down work environments. Existing Raycast options also tend to require Raycast Pro, Obsidian plugins, or a capture flow that does not match how you use your vault.

This repo keeps the path simple:

• It runs from Raycast, no browser extension needed.
• It writes directly to your vault folder.
• The full clipper can use your own LLM.
• The read-later command has no LLM step. It just saves the link.

---

What Each Script Does

Save to Vault

save-to-vault.py is the full clipper.

1. Reads the active tab URL from Chrome or Safari, with clipboard fallback.
2. Detects tweets, X threads, X articles, YouTube videos, and webpages.
3. Extracts the source content. For YouTube, it tries to pull the transcript first and falls back to the video description.
4. Writes a stub note immediately.
5. Runs AI enrichment in the background for summary, key points, tags, and type.
6. Saves the final note to your configured output folder.

Before it fetches anything, the script checks existing saved notes and pending captures for the same source URL. It normalizes common variants like YouTube timestamps, youtu.be links, and tracking parameters, so a title change should not create a second note for the same source.

The saved note keeps the full article body under ## Article or ## Full Content, and YouTube captures keep the transcript under ## Transcript when captions are available. The article_max_chars setting only caps what gets sent to the LLM, not what gets stored in Obsidian.

Read Later to Daily

read-later.py is intentionally small.

1. Reads the active tab URL and title from Chrome or Safari, with clipboard fallback.
2. Cleans the title. For X/Twitter links, it trims long tweet text into a short title instead of dumping the whole post.
3. Appends one bookmark bullet under ## Notes in today's daily note.
4. Creates the daily note if it does not exist yet.
5. De-dupes by URL, so hitting the hotkey twice does not spam the note.

Example output:

- [b] [Bill Guo: designing in code be like...](<https://x.com/billguo/status/123>)

---

Requirements

• macOS
• Raycast
• Python 3.9+
• An Obsidian vault
• Chrome or Safari
• For save-to-vault.py only: Ollama, OpenAI, Anthropic, or another OpenAI-compatible endpoint

read-later.py uses only the Python standard library.

---

Installation

1. Clone the repo

git clone https://github.com/your-username/raycast-vault-clipper.git
cd raycast-vault-clipper

2. Configure

cp config.example.json config.json

Open config.json and set your vault path and browser. If you plan to use the full clipper, also set your LLM provider.

3. Add the scripts to Raycast

• Open Raycast settings.
• Go to Extensions > Scripts.
• Click + > Add Scripts Directory.
• Select this repo folder.
• Search for Save to Vault and assign a hotkey.
• Search for Read Later to Daily and assign a second hotkey.

That is it. No Obsidian plugin, no browser extension.

---

Dependencies

read-later.py has no third-party dependencies.

save-to-vault.py bootstraps its own virtual environment the first time you run it. On first run you will see:

First-run setup: creating venv and installing dependencies...
(this takes 10-30 seconds, only happens once)

That creates .venv/ next to the script and installs requests, trafilatura, markdownify, and youtube-transcript-api. Every run after that uses the local venv.

To warm it up before assigning a hotkey:

./save-to-vault.py

The first run may complain about no URL if you run it from the terminal. That is fine. The venv is ready after that.

If you ever need to rebuild the venv, delete .venv/ and run save-to-vault.py again.

---

Configuration

All settings live in config.json, which is gitignored.

Key	Default	Description
vault_path	~/Documents/Obsidian Vault	Path to your Obsidian vault
output_folder	Sources/clips	Where full saved notes are written
browser	chrome	Browser to read from: chrome or safari
llm.provider	ollama	ollama, openai, or anthropic
llm.model	llama3.2:3b	Model name
llm.base_url	http://localhost:11434	API base URL
llm.api_key	""	API key, not needed for Ollama
capture.max_content_chars	3000	Max short-content characters sent to the LLM
capture.article_max_chars	16000	Max article characters sent to the LLM
capture.tag_count	5	Number of tags to generate
capture.filename_format	title	title, date-title, or title-date

read-later.py only needs vault_path and browser.

LLM Provider Setup

For Ollama:

brew install ollama
ollama pull llama3.2:3b

"llm": {
  "provider": "ollama",
  "model": "llama3.2:3b",
  "base_url": "http://localhost:11434",
  "api_key": ""
}

For OpenAI or an OpenAI-compatible endpoint:

"llm": {
  "provider": "openai",
  "model": "gpt-4o-mini",
  "base_url": "https://api.openai.com",
  "api_key": "sk-..."
}

For Anthropic:

"llm": {
  "provider": "anthropic",
  "model": "claude-haiku-4-5-20251001",
  "base_url": "",
  "api_key": "sk-ant-..."
}

---

Usage

Save a Full Note

1. Open a page in Chrome or Safari.
2. Run Save to Vault from Raycast.
3. A stub note lands in your configured output folder.
4. The background process fills in the AI summary, key points, tags, and final frontmatter.

If no browser tab is available, copy a URL first. The script will use the clipboard as a fallback.

Save a Read-Later Link

1. Open the page, post, or tweet you want to save.
2. Run Read Later to Daily from Raycast.
3. The script appends one [b] bookmark bullet under ## Notes in Journal/Daily/YYYY-MM-DD.md.

You can also run it from the terminal:

./read-later.py --url https://example.com/post --title "Example Post"

For testing without writing to the vault:

./read-later.py --url https://example.com/post --title "Example Post" --dry-run

---

Maintenance Commands

Sweep Pending Saves

If your laptop sleeps or the LLM is down, a pending sidecar may stay in <output_folder>/_pending/. Retry all pending saves with:

./save-to-vault.py --sweep

Fix Processed Notes Missing Summaries

If a saved note is marked processed but no matching downstream summary exists:

./save-to-vault.py --fix-orphans

Debug Browser Detection

X changes its DOM often. If a tweet, thread, or X article clips in the wrong shape, run:

./diagnose-tab.py

It prints the selectors and content lengths the clipper can see in the active tab. That usually tells you whether the issue is detection, extraction, or LLM truncation.

---

Output Examples

Read Later

## Notes
- [b] [Example Post](<https://example.com/post>)

Full Save

---
title: "Punks vs Rude Boys Subcultures"
type: clip
subtype: resource
source: https://x.com/timecaptales/status/1736190618320158819
author: "Time Capsule Tales"
tags:
  - subcultures
  - music-history
  - fashion
date: 2026-03-16
enriched: true
processed: false
domain: subcultures
---

# Punks vs Rude Boys Subcultures

## Summary

A concise AI summary goes here.

## Tweet

> Source content goes here.

## Notes

---

[[subcultures]]  [[music-history]]  [[fashion]]

---

Contributing

PRs welcome. The scripts are intentionally small so they stay easy to inspect, debug, and adapt to a specific vault workflow.