Research Assistant — AI-Powered Deep Research Agent

An OpenClaw agent that conducts thorough web research, verifies sources, and delivers comprehensive reports with citations.

This agent acts as your personal research analyst, capable of:

• Deep web research using browser automation
• Multi-source verification and fact-checking
• Academic-quality citations and bibliographies
• Structured report generation
• Source credibility assessment
• Research project tracking across sessions

What It Does

The Research Assistant transforms vague questions into well-researched, cited reports:

Example workflow:

You: Research the current state of quantum computing applications

Agent:
1. Searches academic sources, industry publications, and news
2. Uses browser automation to read full articles (not just snippets)
3. Verifies claims across multiple sources
4. Assesses source credibility (peer-reviewed > industry blog > random site)
5. Generates a structured report with:
   - Executive summary
   - Key findings by category
   - Methodology notes
   - Full bibliography with links
6. Saves research to memory/research-YYYY-MM-DD.md for future reference

Key Features

🔍 Deep Research Capabilities

• Browser automation — Reads full articles, not just search snippets
• Multi-source verification — Cross-references claims across 3+ sources
• Academic search — Prioritizes peer-reviewed sources when available
• Credibility scoring — Evaluates source reliability

📚 Citation & Documentation

• Automatic citations — APA, MLA, or Chicago format
• Link preservation — All sources linked in bibliography
• Quote extraction — Pulls key quotes with proper attribution
• Methodology transparency — Documents search process

🗂️ Project Management

• Research tracking — Maintains research projects across sessions
• Progressive refinement — Builds on previous research
• Structured outputs — Markdown reports, JSON data exports
• Memory integration — Recalls past research to avoid duplication

⚡ Intelligent Processing

• Claude Sonnet 4 — Balanced speed and analytical depth
• Medium thinking budget — 5,000 tokens for complex analysis
• Semantic search — Find related research in memory

Quick Start

Installation

hatchery run research-assistant

You'll be prompted for required configuration.

Configuration

Create a .env file:

# Required: Anthropic API key for Claude
ANTHROPIC_API_KEY=sk-ant-...

# Required: Messaging platform (choose one)
TELEGRAM_BOT_TOKEN=your_telegram_bot_token
# DISCORD_BOT_TOKEN=your_discord_bot_token
# SLACK_BOT_TOKEN=xoxb-your-slack-token

# Optional: Citation format preference
CITATION_FORMAT=APA  # APA, MLA, or Chicago (default: APA)

# Optional: Source credibility threshold (1-10)
MIN_SOURCE_CREDIBILITY=6  # Skip sources rated below this (default: 5)

First Research Request

Telegram/Discord:

Research the impact of AI on software development productivity

Expected output:

• Structured report (5-10 paragraphs)
• 5-10 cited sources
• Executive summary
• Key findings
• Methodology notes
• Full bibliography

Usage Examples

Academic Research

Research peer-reviewed studies on intermittent fasting and metabolic health from the last 5 years

Agent will:

• Prioritize PubMed, Google Scholar, academic journals
• Focus on recent peer-reviewed studies
• Summarize methodology and findings
• Provide full citations

Market Research

Research the competitive landscape for AI code editors

Agent will:

• Identify major players (Cursor, GitHub Copilot, etc.)
• Compare features and pricing
• Analyze market positioning
• Cite industry reports and company websites

Fact-Checking

Verify: "95% of startups fail within the first 5 years"

Agent will:

• Find original source of the claim
• Cross-reference with authoritative data (SBA, academic studies)
• Assess accuracy and context
• Provide corrected statistic if needed

Technology Deep-Dive

Research how Stripe's payment infrastructure works

Agent will:

• Read Stripe's technical documentation
• Find engineering blog posts and talks
• Analyze third-party technical analyses
• Summarize architecture and key innovations

How It Works

Research Workflow

1. Query Analysis

   • Breaks down your question into sub-queries
   • Identifies key terms and concepts
   • Determines appropriate source types

2. Multi-Source Search

   • web_fetch: Quick lookups for known URLs
   • browser: Full article reading for complex content
   • Searches 5-15 sources depending on query complexity

3. Source Evaluation

   • Peer-reviewed (10/10) > Government/Academic (9/10) > Industry (7/10) > News (6/10) > Blogs (4/10)
   • Filters out sources below credibility threshold
   • Checks publication dates for timeliness

4. Synthesis

   • Cross-references claims
   • Identifies consensus and disagreements
   • Extracts key quotes
   • Structures findings logically

5. Report Generation

   • Executive summary (2-3 sentences)
   • Main findings (organized by theme)
   • Methodology (what was searched, what was found)
   • Bibliography (full citations with links)

6. Memory Storage

   • Saves to memory/research-YYYY-MM-DD.md
   • Indexes findings for future semantic search
   • Tracks research projects

Session Architecture

Main session (operator):

• Full browser automation access
• Can save research to memory
• All tools available

DM sessions (disabled by default):

• This template is designed for operator use only
• Enable DM mode if you want to allow approved users to request research
• DM users get web_fetch (no browser automation for security)

Group sessions (disabled):

• Not recommended — research is too resource-intensive for public groups

Customization

Citation Formats

Edit TOOLS.md to set your preferred format:

## Citation Preferences
- Format: APA  # or MLA, Chicago
- Include DOIs: yes
- Include access dates: no

Source Priorities

Edit AGENTS.md under "Source Credibility" to adjust:

- Peer-reviewed journals: 10/10
- Government data (.gov): 9/10
- Academic institutions (.edu): 9/10
- Industry leaders (Stripe, OpenAI blogs): 7/10
- Reputable news (NYT, WSJ, Reuters): 6/10
- Standard blogs: 4/10
- Unknown sites: 3/10

Research Templates

Add custom research templates to memory/templates/:

# Market Research Template
1. Market size and growth
2. Key players and market share
3. Competitive advantages
4. Pricing models
5. Recent news and trends

Agent will use these templates when requested.

File Structure

workspace/
├── README.md              # This file
├── hatchery.yaml          # Runtime configuration
├── SOUL.md                # Agent personality (analytical researcher)
├── AGENTS.md              # Research workflows and protocols
├── MEMORY.md              # Knowledge base (research methods, source types)
├── IDENTITY.md            # Quick reference card
├── TOOLS.md               # User customization guide
├── .env.example           # Environment variable template
└── memory/
    ├── research-projects.json    # Ongoing research tracking
    ├── research-YYYY-MM-DD.md    # Daily research logs
    └── templates/                # Custom research templates

Troubleshooting

"Browser tool failed"

Cause: Playwright browser not installed or crashed Fix:

# Restart the agent — it will reinstall Playwright
hatchery restart research-assistant

"Too many sources, slow response"

Cause: Complex query requires reading many pages Fix:

• Be more specific in your research question
• Or adjust MAX_SOURCES=5 in .env to limit depth

"Low credibility sources"

Cause: Topic has limited authoritative sources Fix:

• Lower MIN_SOURCE_CREDIBILITY threshold
• Or ask agent to "research with any available sources"

"Citations incomplete"

Cause: Paywalled or JavaScript-heavy sites Fix:

• Agent will note when full text unavailable
• Provide direct links to PDFs if you have them

Advanced Usage

Research Projects

Track multi-session research:

Start a research project on "AI Safety Governance"

Agent creates memory/projects/ai-safety-governance.json and tracks:

• Research questions
• Sources reviewed
• Key findings
• Next steps

Continue later:

Continue research on AI Safety Governance — focus on EU regulations

Comparative Analysis

Compare research on Topic A vs Topic B

Agent will:

• Research both topics separately
• Generate comparison table
• Highlight similarities and differences

Longitudinal Studies

Research "remote work productivity" and compare findings from 2019 vs 2024

Agent filters sources by date range and compares evolution.

Privacy & Ethics

• No data collection: All research stored locally in your workspace
• Respects robots.txt: Won't scrape sites that prohibit it
• Rate limiting: Respects site rate limits (won't DOS)
• Attribution: Always cites sources properly
• Paywalls: Won't attempt to bypass paywalls

Cost Estimate

Typical research request:

• Simple query (5 sources): ~$0.10-0.20
• Medium query (10 sources): ~$0.30-0.50
• Deep research (15+ sources): ~$0.50-1.00

Costs from:

• Claude API calls (Sonnet 4)
• Browser automation overhead
• Medium thinking budget

Contributing

Ideas for enhancements:

• [ ] PDF extraction and analysis
• [ ] Integration with Zotero/Mendeley
• [ ] Export to Notion/Obsidian
• [ ] Scheduled research monitoring (track topic over time)
• [ ] Multi-language research

License

MIT License — Free to use and modify.

---

Built with OpenClaw — The self-hosted AI agent runtime.

Questions? Check out the OpenClaw docs or ask in the community Discord.