Quick Start

Run a literature search and produce AI-grounded paper summaries in about 20 minutes.

Research Framework Quickstart

Prompt-first procedure: Describe the outcome you want in your agent conversation. The agent should select and load the appropriate AIWG assets, explain material changes, request any needed approval, and report verification evidence. Exact commands and flags appear only in the CLI reference.

First time using AIWG? Begin with Install, Connect, and Verify. This guide assumes AIWG is already installed, `all` is deployed for your provider, and the agent-owned context regeneration procedure has connected the agent to this project.

Run a literature search and produce AI-grounded paper summaries in about 20 minutes.

Installation

Use AIWG to complete this documented outcome: Installation
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

After installation, eight agents are deployed to your project and the `.aiwg/research/` directory structure is created.

Your First Literature Review

Step 1: Discover Papers

Search using natural language:

Use AIWG to complete this documented outcome: Search using natural language
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

Output files:

  • `.aiwg/research/discovery/search-results-{timestamp}.json` — All results with relevance scores
  • `.aiwg/research/discovery/search-strategy.md` — Search strategy documentation (for PRISMA compliance)
  • `.aiwg/research/discovery/gap-report-{timestamp}.md` — Under-researched topics detected
  • `.aiwg/research/discovery/acquisition-queue.json` — Papers ready for download

The discovery agent ranks results by: relevance (40%), citation count (30%), venue quality (20%), recency (10%).

To include citation network traversal (papers citing your results and papers they cite):

Use AIWG to complete this documented outcome: To include citation network traversal (papers citing your results and papers they cite)
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

Step 2: Acquire Papers

Download papers from the acquisition queue:

Use AIWG to complete this documented outcome: Download papers from the acquisition queue
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

This downloads PDFs, assigns REF-XXX identifiers, validates FAIR compliance, and records SHA-256 checksums. Expect a 90%+ success rate; paywalled papers are flagged rather than skipped silently.

For a paywalled paper you have locally:

Use AIWG to complete this documented outcome: For a paywalled paper you have locally
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

For a single paper by its ID:

Use AIWG to complete this documented outcome: For a single paper by its ID
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

If some downloads failed:

Use AIWG to complete this documented outcome: If some downloads failed
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

Step 3: Summarize Papers

Use AIWG to complete this documented outcome: Step 3: Summarize Papers
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

For each acquired PDF, the documentation agent: 1. Extracts text (with OCR for scanned papers if needed) 2. Produces a 1-sentence, 1-paragraph, and 1-page summary 3. Extracts structured data: claims, methods, datasets, key findings 4. Validates all claims against the source text 5. Creates a Zettelkasten-style literature note

Output for each paper:

  • `.aiwg/research/knowledge/summaries/REF-XXX-summary.md`
  • `.aiwg/research/knowledge/extractions/REF-XXX-extraction.json`
  • `.aiwg/research/knowledge/notes/REF-XXX-literature-note.md`

For a single paper:

Use AIWG to complete this documented outcome: For a single paper
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

Step 4: Review the Gap Analysis

cat .aiwg/research/discovery/gap-report-latest.md

The gap report identifies topic clusters with few papers — potential research opportunities or areas where your search terms need refinement.

Complete Workflow: 4 Commands

Use AIWG to complete this documented outcome: Complete Workflow: 4 Commands
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

Typical results for 50 papers: ~45 acquired (90%), ~3-4 hours of reading replaced by 5-minute summaries per paper.

Citation Management

Format a citation for a specific paper:

Use AIWG to complete this documented outcome: Format a citation for a specific paper
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

The citation agent supports 9000+ styles. For systematic reviews, it can also build citation networks showing which papers cite each other.

Quality Assessment

Assess a paper's evidence quality:

Use AIWG to complete this documented outcome: Assess a paper's evidence quality
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

Produces a GRADE assessment (High/Moderate/Low/Very Low) with justification. Useful for systematic reviews where you need to rate the body of evidence.

Configuration

The framework reads agent config from `.aiwg/research/config/`. Key settings:

Adjust search ranking in `discovery-agent.yaml`:

discovery_agent:
  ranking:
    relevance_weight: 0.40   # Natural language similarity
    citation_weight: 0.30    # Citation count (popularity proxy)
    venue_weight: 0.20       # Journal/conference quality
    recency_weight: 0.10     # How recent the paper is

Set download concurrency in `research-acquisition-agent.yaml`:

acquisition_agent:
  download:
    concurrent_downloads: 5
    timeout_seconds: 60

Configure summarization strictness in `documentation-agent.yaml`:

documentation_agent:
  hallucination:
    enabled: true
    confidence_threshold: 0.9    # Claims below this are flagged for review
    user_review_required: true   # Pause on flagged claims

Troubleshooting

Empty search results: Check API connectivity with the agent-owned research operation. Consider reformulating your query — semantic search works best with natural language rather than keyword combinations.

Rate limit errors: Set a Semantic Scholar API key for higher limits:

export SEMANTIC_SCHOLAR_API_KEY="your-api-key"

The framework automatically falls back to arXiv and CrossRef when the primary API is unavailable.

Hallucination detected: Review the flagged content in `.aiwg/research/knowledge/logs/REF-XXX-hallucination-detected.log`. Re-run with the strict-validation option to regenerate with tighter checks.

Scanned PDFs with poor text extraction: Add the ocr option flag:

Use AIWG to complete this documented outcome: Scanned PDFs with poor text extraction: Add the ocr option flag
Have it inspect the current state, explain the plan, ask before material
changes, and report the result with verification evidence.

References

  • `@$AIWG_ROOT/agentic/code/frameworks/research-complete/docs/overview.md` — Framework overview
  • `@$AIWG_ROOT/agentic/code/frameworks/research-complete/agents/` — Agent definitions