Academics building open‑source agents for academic and research work

kkaleb1 pts0 comments

Agents4Academia

Learn together · Build together

Agents4Academia

A community of researchers building and understanding<br>agentic tools for academic work.

GitHub

Discord

@agents4academia

Bluesky

literature<br>discovery

knowledge<br>organisation

coding

reproducibility

review

teaching & learning

administration

= an agent

AI agents are changing what individuals can build: a recurring frustration<br>or a highly specific workflow can become a working tool in a fraction of<br>the time it used to take. Across academia, people are already<br>experimenting — literature discovery, knowledge organisation,<br>reproducibility, review, teaching, administration — but mostly in<br>isolation, with lessons rarely shared. Agents4Academia brings these<br>efforts together: sharing practical experience, identifying common needs,<br>and developing open, reusable tools shaped by researchers themselves.

Our goals are simple:

Learn together where agents add real value, where their limitations lie, and how best practices should evolve as the technology improves.

Build together open and reusable tools for shared academic needs, adapting them over time to new capabilities and ways of working.

We welcome researchers and collaborators from all institutions and<br>disciplines to contribute to this shared effort. Our<br>principles guide the work.

Our first cohort (Oxford–Singapore Hackathon, June 2026)

Our first cohort launched with a two-week hackathon (14–26 June 2026),<br>running in parallel across Oxford Statistics, NUS, and NTU with generous<br>support from Anthropic. We deliberately didn't pre-assign projects:<br>participants identified concrete bottlenecks in their own scientific<br>workflows and self-organised into small teams around them, so the work<br>reflects researchers' real, felt needs rather than an imposed agenda.

The hackathon shipped five open-source agents spanning the research lifecycle:

Discover<br>Prior<br>auditable knowledge graphs from primary literature

Organise<br>UReKA<br>Zotero, Notion, Obsidian & arXiv, unified

Reproduce<br>Benchmark-Replicator<br>clean, runnable implementations of prior papers

Verify<br>RefWarden<br>do citations exist, and do they support the claim?

Review<br>Auto-Reviewer<br>claim–evidence maps, novelty and rigor checks

While building side by side, the teams repeatedly encountered the same<br>infrastructure needs. We therefore extracted them into<br>academia-core :<br>shared components for full-text ingestion, citation resolution, and<br>evidence-grounded judging that other research agents can build on.

Get involved

Try the tools. Each repository includes setup instructions and examples.

Contribute. Pull requests, evaluations, integrations, and tests on new domains are welcome.

Build on the shared infrastructure. Use academia-core for your own research agents.

Organise a chapter. Run an Agents4Academia hackathon at your institution.

Join the next cohort. We are planning the next hackathon.

Join the conversation. Find us on Discord, X, and Bluesky.

A collaboration across

agents agents4academia together build shared hackathon

Related Articles