Quick start¶
Get JARVIS RD Assistant running on your own machine and analyze your first paper. This is the happy path for Linux or macOS with Docker. For anything non-standard — GPUs, remote or family access, non-interactive installs — see the Deployment guide.
What you get¶
A self-hosted research workspace: a daily ranked feed of new papers, cross-paper question answering with inline citations, and spaced-repetition cards — running on your own hardware with local Ollama models.
Before you run¶
- Docker Engine 24+ with Compose 2.24+ (Docker Desktop on macOS, or WSL2 on Windows)
- ~27–54 GB free disk for the one-time first install (models and images)
- GPU optional — NVIDIA is fastest; CPU works, but a first paper analysis can take 30+ minutes
Install¶
git clone https://github.com/limitcycle-oss/jarvis-rd-assistant.git
cd jarvis-rd-assistant
./setup.sh
setup.sh checks your system, generates its own secrets, pulls the images and the Ollama models for your hardware, waits for the services, and prints one finish-setup link. Open that link in your browser.
First few minutes in the app¶
- Create your administrator — enter an email and choose Create admin & sign in. The first admin needs no email round-trip.
- Add a research topic in the setup wizard so the feed has something to track.
- Add your first paper — open Discover, search by title or keyword, and choose Save to your papers on the result you want; or use Upload PDF to add a file directly.
- Open the paper and choose Analyze Paper to download, parse, and summarize it. On a GPU this takes a few minutes; on CPU, longer.
You now have a running instance with a paper analyzed. The sidebar starts on a short rail of the six places the daily loop visits; Show all features in its footer opens the rest when you want it.
Next steps¶
- Using the app day to day: First sign-in and setup covers accounts, the wizard, inviting others, and adding papers in full.
- Something non-standard? The Deployment guide covers GPUs (NVIDIA, ROCm, Vulkan), remote and family access, macOS specifics, and unattended installs.