Prepare the official project in your own environment, connect the model/search services you choose, run one bounded research question, then inspect the original sources. Keep manual research as a practical first route.
In this guide
Decide whether the setup fits you
The current official setup guide specifies Python 3.12 or later and model/search credentials. Follow that guide for its current dependency instructions. A browser chat subscription and API access are separate account arrangements; confirm what your chosen provider supplies.
Use the manual product-writing route if you prefer to gather source passages in a browser and paste them into your existing AI service.
Developer route: clone, install, and start the app
The repository Getting Started route specifies Python 3.12 or later. The separate pip-package route has its own requirements; do not mix those instructions. These commands launch your own local app. Submitting research uses your configured providers and budget.
Open http://127.0.0.1:8000, check the provider configuration, and run one narrow question when you are ready to use those accounts. On macOS/Linux, create the environment with python3 and use .venv/bin/python for the same dependency and uvicorn commands. The manual product-research workflow remains an option without this setup.
git clone https://github.com/assafelovic/gpt-researcher.git cd gpt-researcher py -3.12 -m venv .venv .\.venv\Scripts\python.exe -m pip install -r requirements.txt # Create a local .env for your chosen providers: # OPENAI_API_KEY=your-own-key # TAVILY_API_KEY=your-own-key # Keep that file private and out of version control. .\.venv\Scripts\python.exe -m uvicorn main:app --host 127.0.0.1 --port 8000
Judge the evidence, not the report's confidence
- Each important product fact has a direct original source for the correct model.
- The report preserves region, plan, compatibility, and other conditions.
- A missing reference or unavailable source becomes a question to investigate.
- A second generated answer does not establish that the first answer is correct.
If the first run does not work
For credential or quota errors, check the configured provider and account. For startup errors, check the project's required Python version and dependency output. If references do not support the claim, remove it from the brief and research it directly.
Model, search, and hosting charges depend on your configuration. ToolFlowKit does not provide a model proxy or pay those accounts.
References
Use it in a task
AI Product Description Workflow for Store Owners: Research, Write, Review
Continue the task
Turn the research output into a decision you can defend by keeping its evidence and uncertainties available.