An AI-orchestrated workspace for biological data analysis.
ABA is a research environment where a biologist and an AI agent — Guide — work side by side on a real analysis project, carrying it from raw data all the way to results and conclusions. You describe what you want in plain language; Guide plans the analysis, runs the code, and produces results you can inspect, revise, and build on. Guide is powered by a language model of your choice (such as Claude).
Unlike a chatbot bolted onto a notebook, ABA keeps your work as structured, typed, persistent objects — datasets, analyses, figures, findings — each stamped with the provenance of how it was made. Your project is durable: close the browser, come back tomorrow, and everything (data, results, and the reasoning behind them) is where you left it. It's built for long-running research, not one-off questions.
Whatever entity you have open — a Result, a figure, a dataset — Guide already knows what you're looking at. Open an old figure and ask to revise it, or highlight something in the figure or text and just ask.
- Analyze your data by asking. Import a dataset and ask Guide to run quality control, clustering, differential expression, annotation, and more. It picks appropriate methods and runs them for real.
- Work with results, not files. Datasets, analyses, and figures are first-class entities you can pin, revisit, and connect — organized by project, not scattered across folders.
- Trust what you get. Every result carries an execution record — the code, inputs, environment, and the machine it ran on — so any figure is reproducible and reviewable.
- Explore interactively. Rich built-in viewers open your results (e.g. single-cell data in the pagoda3 viewer) directly from a link.
- Run where the compute is. ABA works on your own machine and on any compute you
connect to it — a lab workstation or server reached over SSH, or a Slurm/HPC
cluster — added through a short Settings → Compute flow. Short steps run interactively
on a live kernel; long or heavy steps run as background jobs (an
sbatchjob on a cluster), including large workflow pipelines such as Nextflow / nf-core. Results flow back into your project, and their outputs are kept durably wherever they ran — bring a copy back into your workspace with a click. Offload the compute without changing how you work.
ABA's analysis know-how is organized as a library of recipes that Guide draws on, so its capabilities grow over time without changing the core application.
- A Mac or Linux machine (or access to a Slurm cluster / Open OnDemand).
- An Anthropic API key/subscription or access to some other language model to power the Guide agent.
The installer bootstraps everything else it needs (Python, the analysis environment, and the interface) — you don't have to set those up by hand.
Pick the guide for your setup:
| Setup | Guide |
|---|---|
| Mac (your laptop) | docs/install/mac_personal.md |
| Linux (laptop, workstation, or server) | docs/install/linux_personal.md |
| Slurm cluster (offload jobs to HPC) | docs/install/cluster_personal.md |
| Multi-user cluster (admin setup via Open OnDemand) | docs/install/cluster_open_ondemand.md |
Once installed, ABA opens in your browser; configure Anthropic credentials or a custom LLM, create a project, import your data, and start working with Guide.
- Architecture overview — docs/arch/overview.md
- All documentation — docs/
MIT © 2026 Peter Kharchenko