A sophisticated research tool powered by Minimax M2.1 with interleaved thinking, Exa neural search, and multi-agent orchestration.

• Minimax M2.1 Supervisor: Uses interleaved thinking to maintain reasoning state across multi-step research
• Intelligent Planning: Automatically decomposes research queries into optimized subqueries
• Neural Web Search: Leverages Exa API for high-quality, AI-powered web search
• Comprehensive Reports: Generates detailed research reports with citations and analysis
• CLI Interface: Simple command-line interface with interactive and single-query modes

+-----------------------------------------------+
|            Supervisor Agent                   |
|    (Minimax M2.1 + Interleaved Thinking)      |
+-----------------------------------------------+
                      |
       +--------------+--------------+
       |              |              |
       v              v              v
+------------+ +-------------+ +-----------+
|  Planning  | | Web Search  | | Synthesis |
|   Agent    | |  Retriever  | |  (M2.1)   |
|  (Gemini)  | |             | |           |
+------------+ +-------------+ +-----------+
                      |
                      v
               +------------+
               |  Exa API   |
               +------------+

# Clone and setup
cd deep-research-agent
uv sync

# Configure API keys
cp .env.example .env
# Edit .env with your keys

# Run
uv run python main.py -q "Your research query here"

• Python 3.12+
• uv package manager
• API keys for:
  • Minimax (M2.1 model)
  • OpenRouter (for Gemini)
  • Exa (web search)

1. Install dependencies:

cd deep-research-agent
uv sync

2. Configure environment variables:

cp .env.example .env

Edit .env and add your API keys:

MINIMAX_API_KEY=your_minimax_api_key_here
OPENROUTER_API_KEY=your_openrouter_api_key_here
EXA_API_KEY=your_exa_api_key_here

uv run python main.py

Then enter your research queries at the prompt.

uv run python main.py -q "What are the latest developments in quantum computing?"

uv run python main.py -q "AI trends in 2025" --save

uv run python main.py -q "Climate change solutions" --verbose

The Planning Agent decomposes your query into 3-5 optimized subqueries:

{
  "subqueries": [
    {"query": "quantum computing breakthroughs 2025", "type": "news", "priority": 1},
    {"query": "quantum computing applications cryptography", "type": "auto", "priority": 2}
  ]
}

The Web Search Retriever executes each subquery using Exa:

• Performs neural search for each subquery
• Finds similar content for high-priority results
• Extracts highlights and key information

The Supervisor Agent (Minimax M2.1):

• Maintains reasoning state via interleaved thinking
• Synthesizes comprehensive report with:
  • Table of contents
  • Key takeaways
  • Executive summary
  • Detailed analysis
  • Cited sources

The key innovation: the supervisor preserves ALL content blocks (thinking + text + tool_use) in conversation history. This maintains the reasoning chain across multiple turns for more coherent reports.

deep-research-agent/
├── main.py                    # CLI entry point
├── .env.example               # Environment template
├── pyproject.toml             # Dependencies
└── src/
    ├── agents/
    │   ├── supervisor.py           # Minimax M2.1 supervisor
    │   ├── planning_agent.py       # Query planning
    │   └── web_search_retriever.py # Exa search integration
    ├── tools/
    │   └── exa_tool.py             # Exa API wrapper
    └── utils/
        └── config.py               # Configuration

uv run python main.py -q "What are the latest breakthroughs in artificial general intelligence?"

uv run python main.py -q "What are the emerging trends in electric vehicle adoption?" --save

uv run python main.py -q "What are the most promising approaches to carbon capture technology?" --verbose

Edit system prompt in src/agents/supervisor.py

Edit src/agents/web_search_retriever.py:

• num_results: Results per query (default: 5-10)
• time_period: Date filtering
• content_type: Filter by type (news, research, blog)

• Average query time: 30-60 seconds
• Factors: number of subqueries (3-5), search complexity, LLM response times

MIT License

Built with:

• Minimax M2.1 - Advanced reasoning model
• Exa - Neural web search
• Anthropic SDK - API client
• OpenRouter - LLM routing