Federico De Ponte

Founder, OpenDraft

June 25, 2026

9 min read

Open-Source AI Research Assistant: Self-Host Your Own Paper Writer (2025)

Most AI writing tools are closed boxes: you paste your research topic into someone else's server, pay a monthly subscription, and trust that they handle your data well and cite real sources. There is another path. An open-source AI research assistant lets you read every line of the code, run it on your own machine, and keep full control of your drafts and your keys.

This guide explains what a self-hosted AI research assistant is, why researchers choose one, and how to set up OpenDraft, an MIT-licensed tool that writes thesis-level drafts with verified citations.

What is an open-source AI research assistant?

An open-source AI research assistant is software that helps you search the literature and draft academic writing, where the full source code is published under a permissive license and anyone can inspect, run, or modify it. The key difference from tools like ChatGPT, Jenni, or SciSpace is ownership: you are not renting access to a hosted service, you are running the program yourself.

That distinction matters in three concrete ways:

• Transparency. You can read exactly how the tool searches for papers, how it decides what to cite, and how it verifies sources. Nothing is hidden behind an API.
• Control over data. Your research topic, your draft, and your API keys stay on your machine. No third-party server has to store them.
• No lock-in. There is no subscription to cancel and no vendor that can change pricing or shut down and take your workflow with it. The code keeps working.

Why self-host instead of using a hosted tool?

Hosted AI writing tools are convenient, and for many people that convenience is worth it. Self-hosting makes sense when you care about specific things:

• Cost at volume. Subscriptions add up. With a self-hosted tool you pay only for the model API calls you actually make, and Google Gemini's free tier covers a lot of drafting before you spend anything.
• Privacy. If your topic is sensitive, unpublished, or commercially confidential, running the assistant locally means it never leaves your control.
• Customization. Because the code is yours, you can change the prompts, add your own agents, swap the model, or adjust the citation rules to fit your field.
• Longevity. Open-source tools do not disappear when a startup pivots. You can fork the repository and keep it running indefinitely.

How OpenDraft works under the hood

OpenDraft is an open-source AI research assistant built around a team of 19 specialized agents rather than a single model call. Instead of asking one model to do everything at once, the work is split into stages, so each step is focused and easier to verify:

• Research phase — agents search across academic databases covering 200M+ works to find relevant literature for your topic.
• Writing phase — agents draft structured sections (introduction, methodology, discussion) at the academic level you choose.
• Citation phase — a dedicated verification step checks every source against CrossRef, Semantic Scholar, and arXiv. Anything that cannot be confirmed to exist is removed, which is how OpenDraft avoids the fabricated references that plague single-model tools.

The engine is Python 3.10+, and export runs through Pandoc, XeLaTeX, and python-docx, so you can produce PDF, Word, or LaTeX output. Everything is published under the MIT license, which means you can use it personally, in research, or commercially with no restrictions.

Setting up your self-hosted research assistant

The setup is a standard clone-and-run for anyone comfortable with a terminal:

1. Make sure you have Python 3.10 or higher and Node installed.
2. Clone the OpenDraft repository from GitHub.
3. Install the Python dependencies with pip install -r requirements.txt.
4. Install the front-end dependencies: cd app && npm install.
5. Add your own model API key. Google Gemini has a free tier that is enough for most users to start drafting at no cost.
6. Run npm run dev and generate your first draft locally.

Prefer the command line? OpenDraft also ships an npm CLI (opendraft) so you can generate drafts directly from your terminal without the web interface.

Honest limitations

Open source does not mean magic. OpenDraft is an experiment in multi-agent research drafting, not a guarantee of correctness. You should always read the output critically, check that the verified citations actually support the claims around them, and treat the draft as a strong starting point rather than a finished, submission-ready paper. The advantage of a self-hosted, transparent tool is precisely that you can see where it might go wrong and fix it.

Open source vs. hosted: which should you pick?

If you want zero setup and are happy to use a managed service, a hosted research tool is the faster route. If you value transparency, data control, no subscription, and the ability to modify the tool, a self-hosted open-source assistant like OpenDraft is the better fit. Many researchers use both: a hosted tool for quick exploration and a self-hosted one for sensitive or high-volume work.

Want a hosted version with the same verified-citation engine and nothing to install? The team behind OpenDraft also runs OpenPaper, which gives you the same approach as a managed service.

Related Resources

• Multi-Agent AI for Research: How 19 Agents Work Together
• AI Citation Verification: How OpenDraft Prevents Hallucinated References
• Free AI Research Tools for 2025
• ChatGPT vs OpenDraft: Which AI Tool for Academic Research?
• OpenDraft vs Jenni AI vs Elicit: Best AI Research Tools Compared

About the Author: Written by Federico De Ponte, creator of OpenDraft. OpenDraft is open source under the MIT license; you can read every line of the code on GitHub.