insightflow 0.1.1

A research automation tool that searches and analyzes web content, papers, blogs, and social media to generate comprehensive reports for given queries

insightflow

Japanese / 日本語

Automated research tool that searches and analyzes web content, papers, blogs, and social media to generate comprehensive reports for any query.

Leverages multiple LLM providers (OpenAI, Google, Perplexity, xAI) via OpenRouter to automate multi-perspective research on any topic.

How It Works

Topic
  |
[1] Aspect Extraction (openai/gpt-4.1-mini)
  |
  +-- Aspect A --> [2] Parallel Web Search (perplexity/sonar-reasoning-pro) --> Report A
  +-- Aspect B --> [2] Parallel Web Search                                  --> Report B
  +-- Aspect C --> [2] Parallel Web Search                                  --> Report C
                                                                                 |
                                                            [3] Report Synthesis (google/gemini-3-flash-preview)
                                                                                 |
                                                                    Final Report (Markdown + Citations)

1. Aspect Extraction - Identifies key perspectives of the topic using an LLM
2. Parallel Search - Searches each aspect concurrently via web-connected models
3. Report Synthesis - Merges all findings into a single, cited Markdown report

Prerequisites

• Python 3.12+
• OpenRouter API key

Installation

pip install insightflow

Extras

Extra	Description	Command
cli	CLI interface (Typer)	pip install "insightflow[cli]"
api	REST API server (FastAPI)	pip install "insightflow[api]"
mcp	MCP server (Claude Code, etc.)	pip install "insightflow[mcp]"
all	All extras	pip install "insightflow[all]"

Setup

Set your OpenRouter API key as an environment variable:

export OPENROUTER_API_KEY="sk-or-v1-..."

Or create a .env file in your project root:

OPENROUTER_API_KEY=sk-or-v1-...

Usage

Python Library

import asyncio
from insightflow.core import research
from insightflow.models import LLMConfig

report = asyncio.run(research(
    topic="Recent trends in quantum computing",
    aspect_model=LLMConfig(model="openai/gpt-4.1-mini"),
    search_model=LLMConfig(model="perplexity/sonar-reasoning-pro"),
    report_model=LLMConfig(model="google/gemini-3-flash-preview"),
))

print(report.content)       # Markdown report
print(report.citations)     # List of citations
print(report.metadata)      # Elapsed time, model used, etc.

Individual functions are also available:

from insightflow.core import extract_aspects, search, compose
from insightflow.models import LLMConfig

# Aspect extraction only
aspects = asyncio.run(extract_aspects(
    topic="ML optimization techniques",
    config=LLMConfig(model="openai/gpt-4.1-mini"),
))

# Single search query
result = asyncio.run(search(
    query="Python packaging best practices",
    config=LLMConfig(model="perplexity/sonar-reasoning-pro"),
))

CLI

pip install "insightflow[cli]"

# Full research
insightflow research "Recent trends in quantum computing"

# Aspect extraction only
insightflow aspects "ML optimization techniques"

# Single search
insightflow search "Python packaging best practices"

# With options
insightflow research "AI safety" \
  --language english \
  --max-aspects 3 \
  --search-model perplexity/sonar-pro \
  --json \
  -o report.json

Run insightflow --help for all available options.

REST API Server

pip install "insightflow[api]"
python -m uvicorn insightflow.interfaces.api:app

Swagger UI is available at http://localhost:8000/docs.

curl -X POST http://localhost:8000/research \
  -H "Content-Type: application/json" \
  -d '{"topic": "quantum computing", "language": "english"}'

MCP Server (Claude Code Integration)

Use insightflow as an MCP tool in Claude Code so that Claude can autonomously run research.

1. Install

pip install "insightflow[mcp]"

2. Add to Claude Code config

Add to ~/.claude.json (global) or .mcp.json in your project root:

{
  "mcpServers": {
    "insightflow": {
      "command": "python",
      "args": ["-m", "insightflow.interfaces.mcp"],
      "env": {
        "OPENROUTER_API_KEY": "sk-or-v1-..."
      }
    }
  }
}

With uvx (no local install required):

{
  "mcpServers": {
    "insightflow": {
      "command": "uvx",
      "args": ["--from", "insightflow[mcp]", "python", "-m", "insightflow.interfaces.mcp"],
      "env": {
        "OPENROUTER_API_KEY": "sk-or-v1-..."
      }
    }
  }
}

3. Use from Claude Code

Once configured, Claude Code recognizes the research tool. Ask Claude to research a topic in conversation and it will call insightflow automatically.

Tool parameters:

Parameter	Type	Required	Default	Description
topic	string	Yes	-	Research topic
aspect_model	string	No	openai/gpt-4.1-mini	Model for aspect extraction
search_model	string	No	perplexity/sonar-reasoning-pro	Model for search
report_model	string	No	google/gemini-3-flash-preview	Model for report generation
language	string	No	japanese	Output language
max_aspects	integer	No	5	Maximum number of aspects
concurrency	integer	No	3	Maximum concurrent searches

Configuration

Configurable via environment variables or .env file:

Variable	Required	Default	Description
OPENROUTER_API_KEY	Yes	-	OpenRouter API key
DEFAULT_ASPECT_MODEL	No	openai/gpt-4.1-mini	Default aspect extraction model
DEFAULT_SEARCH_MODEL	No	perplexity/sonar-reasoning-pro	Default search model
DEFAULT_REPORT_MODEL	No	google/gemini-3-flash-preview	Default report generation model
DEFAULT_LANGUAGE	No	japanese	Default output language
DEFAULT_MAX_ASPECTS	No	5	Default number of aspects
DEFAULT_CONCURRENCY	No	3	Default concurrent searches

Development

git clone https://github.com/sync-dev-org/insightflow.git
cd insightflow
uv sync --all-extras

# Test
uv run pytest

# Lint
uv run ruff check src/
uv run ruff format src/

License

MIT License - See LICENSE for details.