online-research

name	online-research
description	Discover and download academic PDFs using DOIs, arXiv IDs, or paper titles. Downloaded PDFs are automatically processed by the PDF monitor service.
tools	["download_pdf","locate_pdf"]

Online Research & PDF Discovery

Discover and download academic PDFs from various sources using DOIs, arXiv IDs, or paper titles.

Core Capabilities

This skill enables you to:

• Locate PDFs from DOIs, arXiv IDs, or paper titles
• Download PDFs directly to the vault (auto-processed by monitor service)

Note: For web search, use Letta's built-in web_search tool instead of Thoth's deprecated web search. Downloaded PDFs are automatically processed by the PDF monitor service - no manual processing needed.

Reading External Content

For reading web articles, blog posts, and documentation outside your knowledge base:

Tool	Use For
Letta's fetch_webpage	Read external web content (blogs, docs, articles)
read_full_article	Read papers in your knowledge base
download_pdf	Save papers for later processing

Iterative Learning with External Sources:

1. Use web_search to find relevant web content
2. Use fetch_webpage to read articles fully
3. Learn from the content, identify knowledge gaps
4. Keep reading more sources until you understand the topic
5. For academic papers, use download_pdf to add them to your knowledge base

Tools Overview

Tool	Purpose	When to Use
locate_pdf	Find PDF URLs from identifiers	Have DOI/arXiv ID, need download link
download_pdf	Download PDF to vault	Found a paper URL, want to save it

When to Use This Skill

Use online research when user asks for:

• "Download this paper from arXiv"
• "Get the PDF for DOI: 10.xxxx/xxxxx"
• "Find and download this paper"
• "Add this paper to my knowledge base"

For web search: Use Letta's built-in web_search tool to find papers online. Then use this skill to download them.

Don't use when:

• User asks about papers already in knowledge base (use knowledge-base-qa or deep-research skills)
• Simple citation lookup (use citation tools)

Standard Research Workflow

1. Discover Papers Online (Use Letta web_search)

Use Letta's built-in web_search tool to find papers:

• Add year filters: "reinforcement learning 2024 2025"
• Use specific venues: "NeurIPS 2024 attention mechanisms"
• Include file type: "transformers architecture filetype:pdf"
• Target preprint servers: "site:arxiv.org memory augmented"

2. Locate PDF Sources

# If you have a DOI
locate_pdf(
  doi="10.1145/3534678.3539147",
  title="[Paper Title]"
)

# If you have an arXiv ID
locate_pdf(
  arxiv_id="2401.12345",
  title="[Paper Title]"
)

# If you only have a title
locate_pdf(
  title="Attention Is All You Need"
)

Returns:

• Direct PDF URL
• Source (arXiv, ACM, IEEE, DOI resolver, etc.)
• Accessibility status
• Alternative links if primary fails

3. Validate Before Downloading

# For multiple papers, validate sources first
validate_pdf_sources(
  sources=[
    {"url": "https://arxiv.org/pdf/2401.12345.pdf"},
    {"doi": "10.xxxx/xxxxx"},
    {"arxiv_id": "2402.56789"}
  ]
)

Validation checks:

• URL accessibility
• Content-Type verification
• File size reasonableness
• Redirect handling

4. Download PDFs

# Download single PDF
download_pdf(
  source="https://arxiv.org/pdf/2401.12345.pdf"
)

# Or use DOI
download_pdf(
  source="10.1145/3534678.3539147"
)

# Or use arXiv ID
download_pdf(
  source="2401.12345"
)

# Custom filename (optional)
download_pdf(
  source="https://example.com/paper.pdf",
  filename="My Custom Paper Name.pdf"
)

Download features:

• Auto-saves to configured vault PDF directory
• Auto-generates filenames from titles
• Supports direct URLs, DOIs, and arXiv IDs
• Progress tracking for large files
• Duplicate detection

5. Automatic Processing

Downloaded PDFs are automatically picked up by the PDF monitor service, which:

• Extracts text and metadata
• Generates embeddings
• Adds papers to the knowledge base
• Integrates with citation network
• Auto-generates tags

No manual processing needed! Just download the PDF and the monitor service handles the rest.

Workflow Examples

Example 1: Find and Download Papers

User: "Find and download the latest papers on memory-augmented transformers"

Step 1: Use Letta's web_search to find papers
web_search(query="memory-augmented transformers 2024 2025 arxiv")
→ Found papers with arXiv links

Step 2: Extract arXiv IDs or DOIs from results

Step 3: Download papers
For each paper:
  download_pdf(source="2401.12345")  # arXiv ID
→ Downloaded and auto-processed

Response: "Downloaded 5 papers on memory-augmented transformers.
They will be automatically processed and added to your knowledge base."

Example 2: Download Specific Paper

User: "Download the paper 'Attention Is All You Need'"

Step 1: Locate the paper
locate_pdf(title="Attention Is All You Need")
→ Found: https://arxiv.org/pdf/1706.03762.pdf

Step 2: Download
download_pdf(source="https://arxiv.org/pdf/1706.03762.pdf")
→ Downloaded to: vault/thoth/papers/pdfs/Attention-Is-All-You-Need.pdf
→ Auto-processing will add it to knowledge base

Response: "Successfully downloaded 'Attention Is All You Need'.
The paper will be automatically processed and added to your knowledge base."

Example 3: Download from DOI

User: "Get me the PDF for DOI: 10.1145/3534678.3539147"

Step 1: Download directly using DOI
download_pdf(source="10.1145/3534678.3539147")
→ Downloaded and auto-processed

Response: "Downloaded paper from DOI 10.1145/3534678.3539147.
The paper will be automatically processed and added to your knowledge base."

Example 4: Research Topic Discovery

User: "What's the latest research on efficient attention mechanisms? Download the top 5 papers."

Step 1: Use Letta's web_search
web_search(query="efficient attention mechanisms 2024 2025 arxiv")
→ Found papers with arXiv links

Step 2: Extract arXiv IDs from results

Step 3: Download top 5 papers
For each arXiv ID:
  download_pdf(source="[arxiv_id]")
→ All downloaded and auto-processed

Response: "Downloaded 5 papers on efficient attention mechanisms.
All papers will be automatically processed and added to your knowledge base."

Advanced Techniques

Smart Filename Generation

# The download_pdf tool auto-generates filenames from:
# 1. Article title (if found in metadata)
# 2. URL path (if title unavailable)
# 3. Hash-based unique name (as fallback)

# You can override with custom filename:
download_pdf(
  source="https://arxiv.org/pdf/2401.12345.pdf",
  filename="MyResearch_ReinforcementLearning_2024.pdf"
)

Error Handling

# If direct download fails, try locating the PDF first
# Attempt primary source
pdf = locate_pdf(doi="10.xxxx/xxxxx")
download_pdf(source=pdf.url)
# If locate fails, try Letta's web_search to find alternative URLs

Source Priority

When locating PDFs, the system tries sources in this order:

1. arXiv - Fast, reliable, open access
2. DOI resolver - Authoritative, may require access
3. Semantic Scholar - Good metadata, open access tracking
4. PubMed Central - Life sciences, open access
5. ACM/IEEE - Conference papers (may require access)

Best Practices

Download Strategy

1. Handle errors: Not all papers are openly accessible
2. Respect limits: Don't overwhelm servers with requests
3. Check duplicates: Tool detects existing files automatically
4. Use arXiv/DOI when available: Most reliable sources

Search Strategy (using Letta web_search)

1. Be specific: Include domain terms, years, venues
2. Use filters: Add "filetype:pdf" or "site:arxiv.org"
3. Iterate: Refine search based on initial results

Common Pitfalls

❌ Don't:

• Search for papers already in knowledge base (use existing tools)
• Use overly broad search queries
• Download papers without checking for duplicates

✅ Do:

• Check knowledge base first with search_articles
• Use specific, targeted search queries with Letta web_search
• Let the download tool handle duplicate detection
• Trust the PDF monitor service for processing

Integration with Other Skills

Online Research → Deep Research

1. Use Letta web_search to find papers
2. Use download_pdf to save them
3. Wait for auto-processing
4. Use read_full_article to deeply read the papers
5. Keep reading related papers to fill knowledge gaps
6. Use deep-research tools to analyze and synthesize

Online Research → Knowledge Base

1. web_search (Letta) → find papers
2. download_pdf → save to vault
3. (auto-processing by monitor service)
4. read_full_article → deeply read the processed papers
5. Now accessible via all knowledge base tools

Learning from External Web Content

1. web_search (Letta) → find web articles, blogs, documentation
2. fetch_webpage (Letta) → read the full content
3. Learn from the article, take notes
4. Identify gaps in understanding
5. Keep reading more sources until you understand

Quick Reference

Fastest Path: arXiv Paper

download_pdf(source="2401.12345")  # Just the arXiv ID
→ Auto-processed by monitor service

Fastest Path: DOI

download_pdf(source="10.1145/xxxxx")
→ Auto-processed by monitor service

Fastest Path: Direct URL

download_pdf(source="https://arxiv.org/pdf/2401.12345.pdf")
→ Auto-processed by monitor service

Discovery Workflow

web_search (Letta) → locate_pdf → download_pdf → (auto-processing)

Summary

This skill empowers you to:

1. Locate PDFs from DOIs, arXiv IDs, or titles
2. Download papers directly to the vault
3. Auto-process via the PDF monitor service

Use this skill as the first step in any research workflow that requires accessing papers not yet in the knowledge base. Once papers are downloaded and processed, switch to deep-research or knowledge-base-qa skills for analysis.