SecondOrder

Deep research that thinks beyond the obvious.

SecondOrder is an AI-powered deep research tool that goes beyond first-pass answers. Where typical search tools retrieve surface-level results, SecondOrder applies second-order thinking — it plans research angles, searches in parallel, scrapes primary sources, and synthesizes everything into a comprehensive, cited report.

Why "SecondOrder"?

In decision-making, first-order thinking asks: "What's the answer?" Second-order thinking asks: "What are the consequences of that answer? What connections am I missing? What does the evidence actually say when you dig deeper?"

Most research tools stop at first-order — they return search snippets and call it done. SecondOrder decomposes your question into multiple angles, investigates each one independently, reads the actual sources, cross-references findings, and produces a report with real citations. It thinks about your question the way a skilled researcher would.

How It Works

You ask a question
       |
       v
  [Orchestrator] — breaks query into 3-5 research angles
       |
       v
  [Search Agents] — parallel web searches via Jina AI
       |
       v
  [Scraper Agents] — extracts full content from top sources via Jina AI Reader
       |
       v
  [Analyzer] — configured OpenRouter model synthesizes everything into a cited report
       |
       v
  Streamed to you in real time via SSE

Every step streams progress — you see the plan form, searches execute, sources get scraped, and the final report writes itself token by token.

Architecture

• Backend: Python + FastAPI, raw asyncio orchestration with OpenRouter (OpenAI-compatible SDK client)
• Search: Jina AI Search
• Scraping: Jina AI Reader
• Models: OpenRouter model IDs (user-selectable, e.g., GPT, Gemini, Llama families)

No agent frameworks. No LangChain. No LlamaIndex. Just Python, asyncio, and OpenRouter-compatible SDK calls.

Quick Start

Prerequisites

• Python 3.12+ (with uv)
• API keys: OpenRouter, Jina AI

LLM Configuration (OpenRouter)

SecondOrder is OpenRouter-only. Configure your key and default model:

# .env
OPENROUTER_API_KEY=sk-or-...
DEFAULT_MODEL=openai/gpt-4o-mini
JINA_API_KEY=jina_...                 # Jina AI API key
SEARCH_PROVIDER=jina                  # jina

Available model examples:

• openai/gpt-4.1 — OpenAI GPT-4.1
• openai/gpt-4o-mini — OpenAI GPT-4o Mini
• google/gemini-2.0-flash-001 — Google Gemini 2.0 Flash
• meta-llama/llama-3.3-70b-instruct — Llama 3.3 70B Instruct
• ... and many more

Check openrouter.io/models for the full list and current pricing.

Backend

cp .env.example .env          # Fill in your API keys
uv pip install -e ".[dev]"
uv run uvicorn app.main:app --reload

Benchmarks

SecondOrder is evaluated against three industry-standard deep research benchmarks:

Benchmark	Source	Tasks	What It Measures
DRACO	Perplexity	100	Rubric-based research quality across domains
ResearchRubrics	Scale AI	2,500+	Expert rubric compliance across 9 domains
DeepSearchQA	Google DeepMind	900	Factual answer extraction (F1 score)

uv run python -m benchmarks.run draco --limit 5
uv run python -m benchmarks.run deepsearchqa --limit 10
uv run python -m benchmarks.run all --limit 3

License

MIT