AI deep research tool that searches the web, reads sources, and synthesizes findings into comprehensive reports.

Works 100% free with Ollama (local LLM) + DuckDuckGo (no API key needed for search). Also supports OpenAI, Google Gemini, Anthropic Claude, and OpenRouter (200+ models).

No Langchain dependency. No paid search APIs required. Just pip install deepworm and go.

pip install deepworm

# Option 1: Google Gemini (recommended, free tier available)
export GOOGLE_API_KEY="AIza..."

# Option 2: OpenRouter (200+ models, free & paid)
export OPENROUTER_API_KEY="sk-or-..."

# Option 3: OpenAI
export OPENAI_API_KEY="sk-..."

# Option 4: Anthropic Claude
export ANTHROPIC_API_KEY="sk-ant..."

# Option 5: Ollama (fully local, no key needed)
# Just install Ollama and pull a model

deepworm "what are the latest advances in quantum computing?"

That's it. DeepWorm will:

1. Search the web for relevant sources
2. Read and analyze each source
3. Generate followup questions
4. Go deeper recursively
5. Produce a final report with citations

# Control research depth and breadth
deepworm "topic" --depth 2 --breadth 3

# Verbose mode (see searches, pages, analysis in real-time)
deepworm "topic" -v

# Save report to file
deepworm "topic" --output report.md

# Auto-polish analysis (readability, quality score, compliance)
deepworm "topic" --polish

# Knowledge graph extraction
deepworm "topic" --graph              # Mermaid diagram
deepworm "topic" --graph stats        # Top connected concepts table

# Full pipeline
deepworm "topic" -v --polish --graph stats

# Choose provider
deepworm "topic" --provider google
deepworm "topic" --provider openrouter --model google/gemini-2.0-flash-001

# Compare topics
deepworm compare "Python" "Rust"

# See all options
deepworm --help

deepworm interactive

Features:

• Arrow-key menu navigation — navigate commands without typing
• /keys — manage API keys interactively (saved to ~/.deepworm_keys)
• /models — browse and switch models with arrow keys
• /set — change depth, breadth, model inline
• /last — view last research report
• /graph — extract knowledge graph from last report
• /polish — run quality analysis on last report
• Command history — arrow up/down to recall previous queries
• Auto-save — all reports saved to ~/.deepworm/reports/

from deepworm import Researcher

researcher = Researcher(provider="openai")
result = researcher.research("your topic")
print(result.report)

Create deepworm.yaml in your project root:

provider: google
model: gemini-pro
depth: 3
breadth: 5
output_dir: ./reports

• Multi-provider LLM support — OpenAI, Google Gemini, Anthropic, OpenRouter (200+ models), Ollama
• Recursive deep research with configurable depth and breadth
• Interactive TUI with arrow-key navigation, command menu, and auto-save
• Knowledge graph extraction — Mermaid, DOT, stats, JSON formats
• Polish pipeline — readability scoring, compliance checks, quality grades (A-F)
• Auto-save reports to ~/.deepworm/reports/ (both CLI and interactive)
• API key management — /keys command saves keys securely to ~/.deepworm_keys
• Markdown reports with citations and source links
• Comparison mode for side-by-side analysis
• Research chains for multi-step investigations
• Built-in retry and rate limiting
• DuckDuckGo search (no API key needed for search)
• Rich terminal output with emoji progress tracking
• Token usage tracking per research
• Extensible plugin system

MIT