npm package discovery and stats viewer.

Discover Tips

  • General search

    [free text search, go nuts!]

  • Package details

    pkg:[package-name]

  • User packages

    @[username]

Sponsor

Optimize Toolset

I’ve always been into building performant and accessible sites, but lately I’ve been taking it extremely seriously. So much so that I’ve been building a tool to help me optimize and monitor the sites that I build to make sure that I’m making an attempt to offer the best experience to those who visit them. If you’re into performant, accessible and SEO friendly sites, you might like it too! You can check it out at Optimize Toolset.

About

Hi, 👋, I’m Ryan Hefner  and I built this site for me, and you! The goal of this site was to provide an easy way for me to check the stats on my npm packages, both for prioritizing issues and updates, and to give me a little kick in the pants to keep up on stuff.

As I was building it, I realized that I was actually using the tool to build the tool, and figured I might as well put this out there and hopefully others will find it to be a fast and useful way to search and browse npm packages as I have.

If you’re interested in other things I’m working on, follow me on Twitter or check out the open source projects I’ve been publishing on GitHub.

I am also working on a Twitter bot for this site to tweet the most popular, newest, random packages from npm. Please follow that account now and it will start sending out packages soon–ish.

Open Software & Tools

This site wouldn’t be possible without the immense generosity and tireless efforts from the people who make contributions to the world and share their work via open source initiatives. Thank you 🙏

© 2026 – Pkg Stats / Ryan Hefner

sourceloop

v0.1.3

Published

NotebookLM-centered research orchestration pipeline

Vulnerabilities

Powered byPowered by Snyk
  • 0High
  • 0Medium
  • 0Low

Links

Git

Maintainers

lteawoolteawoo

Keywords

Readme

SourceLoop

A local-first research runtime for Codex, Claude Code, and Gemini CLI.

SourceLoop helps AI tools work through NotebookLM with a managed browser, repeatable question planning, and local Markdown archives.

It is built for a simple idea:

  • use AI for research
  • keep grounding in real sources
  • keep the workflow reusable
  • keep human ownership over judgment and expression

SourceLoop is not a NotebookLM replacement. It is the workflow and archive layer around NotebookLM-based research.

Why This Exists

Most research workflows break in three places:

  • weak questions
    • the model can answer well only if someone asks the right questions
  • untrusted answers
    • generic AI answers drift, summarize poorly, or hallucinate beyond the source base
  • no durable archive
    • browser-only Q&A disappears into a tab and becomes hard to reuse later

SourceLoop exists to make that workflow repeatable.

It gives your AI tool:

  • a topic model
  • a managed NotebookLM browser path
  • notebook creation or binding
  • source import and source declaration
  • planned question batches
  • local answer and citation archives

The IPO Model

SourceLoop follows a simple Input -> Process -> Output model.

Input

  • local notes, files, transcripts, and URLs
  • a NotebookLM notebook
  • a managed research browser session

Process

  • plan better questions
  • run NotebookLM Q&A against grounded sources
  • capture answers and citations into local files

Output

  • a reusable Markdown research archive
  • question batches you can rerun or extend
  • material that can become memos, articles, scripts, decks, or teaching assets

In short:

Sources
  -> NotebookLM
  -> Planned questions
  -> Grounded answers with citations
  -> Local Markdown archive
  -> Your own memo, deck, article, or presentation

What SourceLoop Gives Your AI Tool

  • topic-first research workspaces
  • managed Chrome setup for NotebookLM work
  • notebook create or notebook bind flows
  • source import for local files and supported remote URLs
  • source declaration when material already exists in NotebookLM
  • default 10-question planning
  • full-batch execution by default unless you explicitly want a partial run
  • local Markdown storage for runs, answers, and citations

Why NotebookLM

SourceLoop uses NotebookLM as the answer engine because NotebookLM is already strong at document-grounded research.

  • Gemini-powered preprocessing
    • upload source material once and work from prepared source context
  • natural-language Q&A
    • go beyond keyword lookup into explanation, synthesis, and comparison
  • multi-source reasoning
    • connect ideas across a notebook worth of source material
  • built-in citations
    • answers come with traceable source references
  • no separate RAG infrastructure
    • no vector database, embedding pipeline, chunking strategy, or retrieval tuning stack to maintain

SourceLoop does not try to replace those strengths. It makes them operational for AI tools.

Markdown-First Output

alt text SourceLoop is designed so the result does not disappear into a browser tab.

Every run is captured as local Markdown artifacts:

  • planned questions
  • NotebookLM answers
  • citation-backed notes
  • run-level archives

This makes the output:

  • easy to reuse
    • turn answers into memos, decks, scripts, articles, or teaching material
  • easy to search
    • keep everything in local files instead of browser history
  • easy to extend
    • use previous runs as source material for later research
  • easy to organize
    • keep an Obsidian-friendly archive that can grow into a real knowledge base
  • easy to version
    • fit naturally into Git-based workflows

SourceLoop does not just help you get answers. It helps you keep them in a form you can build on.

Security Model

SourceLoop recommends using a dedicated SourceLoop-managed Chrome profile for research work.

  • do not use your normal personal Chrome profile as the preferred research path
  • sourceloop chrome launch creates a separate research browser profile
  • after signing in to Google and NotebookLM once, you can keep reusing that research session
  • this keeps research activity separate from your personal cookies, tabs, and extension context
  • if the preferred SourceLoop browser is unavailable, another Chrome session can still be used, but only after explicit user approval

In short:

  • preferred path: a dedicated research browser
  • fallback path: another existing Chrome session
  • fallback use requires explicit user confirmation

Installation

# Install the CLI globally
npm install -g sourceloop

# Go to your project
cd /path/to/your/project

# Initialize SourceLoop
sourceloop init

# Install project-local AI bootstrap
sourceloop init --ai codex
sourceloop init --ai claude
sourceloop init --ai gemini

Usage

Use SourceLoop through your AI tool in plain language.

Start New Research

You can say:

Start research on attention in transformers.

Or:

Start research on attention in transformers and use this video first:
https://www.youtube.com/watch?v=eMlx5fFNoYc

SourceLoop helps the AI tool:

  • prepare a managed research browser
  • create a topic and notebook
  • import the provided sources
  • generate a topic-tailored question batch
  • archive answers and citations locally

Continue Existing Research

You can say:

Continue research from this NotebookLM notebook:
https://notebooklm.google.com/notebook/<real-notebook-id>

SourceLoop helps the AI tool:

  • bind the existing notebook
  • inspect the current research state
  • continue planning or running from there

If You Prefer Direct CLI

Use the Golden Path below.

Golden Path

mkdir my-research-workspace
cd my-research-workspace
sourceloop init --ai codex

sourceloop status --json
sourceloop doctor --json

sourceloop topic create --name "AI agents market" --json

# visible setup step: launch a dedicated research browser,
# sign in yourself, then validate NotebookLM home access
sourceloop chrome launch --name work-chrome
sourceloop attach validate attach-work-chrome

sourceloop notebook-create \
  --name "AI Agents" \
  --topic-id topic-ai-agents-market \
  --attach-target attach-work-chrome \
  --json

# read the returned binding id from JSON or status output
sourceloop ingest ./research-notes.md --topic topic-ai-agents-market

sourceloop notebook-import \
  --notebook <managed-notebook-binding-id> \
  --source-id <source-id> \
  --json

sourceloop notebook-import \
  --notebook <managed-notebook-binding-id> \
  --url "https://youtube.com/watch?v=..." \
  --json

sourceloop plan topic-ai-agents-market --max-questions 10 --json

# or persist AI-authored questions first, then import them into the plan
sourceloop plan \
  topic-ai-agents-market \
  --questions-file ./ai-questions.json \
  --json

sourceloop run <run-id> --json

Operator Defaults

If you use SourceLoop through Codex, Claude Code, Gemini CLI, or another local AI tool, these are the intended defaults:

  • start with sourceloop status --json and sourceloop doctor --json
  • ask for the topic first if the user did not provide one
  • if the user provided a topic but not sources, ask which sources to use before collecting or importing anything
  • mention that planning defaults to 10 questions unless the user wants another count
  • prefer AI-authored topic-specific questions when the operator can generate them, and fall back to the built-in template planner when it cannot
  • prefer sourceloop chrome launch over attaching to an unrelated Chrome profile
  • treat sourceloop attach validate <target> as NotebookLM home validation
  • use --notebook-url only when you specifically need to validate an existing notebook detail page
  • after initial setup, prefer hidden notebook actions by default and use --show-browser only for debugging or when the user wants to watch
  • if the plan was created for 10 questions and the user did not ask for a partial pass, run the planned batch end to end
  • use --limit only for explicit partial runs
  • tell the user briefly that NotebookLM actions can take a bit before waiting
  • if the wait becomes long, ask whether to keep waiting or report the current state
  • if the preferred SourceLoop browser is unavailable, ask whether to continue with the current Chrome or switch back to the SourceLoop browser

Existing Notebook Path

If the notebook already exists and already has sources loaded in NotebookLM:

sourceloop notebook-bind \
  --name "AI Agents" \
  --topic-id topic-ai-agents-market \
  --url "https://notebooklm.google.com/notebook/<real-notebook-id>" \
  --attach-target attach-work-chrome \
  --json

sourceloop notebook-source declare \
  --topic-id topic-ai-agents-market \
  --notebook <notebook-binding-id> \
  --kind mixed \
  --title "AI agents market source set" \
  --json

Partial Runs and Backfills

Use partial execution only when you explicitly want it.

sourceloop run <run-id> --limit 2 --json
sourceloop run <run-id> --from-question <question-id> --json
sourceloop run <run-id> --question-id <question-id> --json
sourceloop import-latest <run-id> --question-id <question-id> --show-browser

Machine-Readable Workflow

Core operator commands support --json:

  • topic create|list|show
  • notebook-create
  • notebook-bind
  • notebook-import
  • notebook-source declare|list|show
  • plan
  • run
  • status
  • doctor

Example:

sourceloop status --json
sourceloop doctor --json
sourceloop topic create --name "AI agents market" --json
sourceloop chrome launch --name work-chrome
sourceloop attach validate attach-work-chrome
sourceloop notebook-create --name "AI Agents" --topic-id topic-ai-agents-market --attach-target attach-work-chrome --json
sourceloop notebook-import --notebook <managed-notebook-binding-id> --url "https://youtube.com/watch?v=..." --json
sourceloop plan topic-ai-agents-market --max-questions 10 --json
sourceloop plan topic-ai-agents-market --questions-file ./ai-questions.json --json
sourceloop run <run-id> --json

Vault Layout

vault/
├─ chrome-targets/
├─ notebook-imports/
├─ notebook-setups/
├─ notebooks/
├─ notebook-sources/
├─ runs/
├─ sources/
└─ topics/

Main result:

vault/runs/<run-id>/
├─ index.md
├─ questions.md
└─ exchanges/
   ├─ q01....md
   ├─ q02....md
   └─ ...

Project Boundary

SourceLoop stops at research packaging and Q&A archive creation.

  • humans choose the topic
  • NotebookLM answers from the bound notebook
  • SourceLoop stores the research trace
  • humans turn that archive into slides, scripts, memos, lessons, or deliverables

This is the intended split:

  • use AI for research
  • use humans for judgment and expression

Current Status

Current focus:

  • topic-first NotebookLM workflow
  • managed notebook setup workflow
  • attached Chrome execution
  • Obsidian-friendly Markdown archive

Still rough around:

  • NotebookLM UI selector stability
  • citation capture fidelity
  • automatic NotebookLM source introspection beyond operator-declared manifests

Development

pnpm install
pnpm build
pnpm test

For local CLI usage:

pnpm install
pnpm build
pnpm link --global

sourceloop init --ai codex
sourceloop --help

Reference docs: