STATUS: Announced. Applications open soon sign up to be notified

A catalyst fund for research software in the age of AI

Supporting the practices that make AI-assisted research software trustworthy.

Grants for the people building, verifying, and teaching research software, and for the norms, tools, methods, and training the community needs to use generative AI well.

What this fund supports

Two tracks, one year, community-set priorities.

Track 01 Coming soon

Catalyst Exploratory Fund

Small, fast grants for work that is ready to begin: writing sprints, tool-building sessions, conference workshops, communities of practice.

Award
up to $50,000
Duration
3–6 months
Total available
to be confirmed
Deadline
late October 2026
Track 02 Planned

Catalyst Impact Fund

Larger, longer awards for work with community-wide reach: building shared tools and infrastructure, studying how practice is changing, delivering training at scale.

Award
up to $200k
Duration
9–12 months
Total available
to be confirmed
Opens
late 2026 / early 2027

What a grant might look like

Examples, not categories.

  1. Agentic verification tools that scaffold software testing and review~$50k
  2. Agent workflows that take on routine maintenance: issue triage, dependency updates, release chores~$30k
  3. A public, searchable collection of case studies from real AI-assisted projects~$25k
  4. A writing sprint to produce a costs-and-benefits framework~$20k
  5. A playbook for research team leaders navigating AI adoption~$15k

Why this fund exists

AI-assisted coding can bypass two checks research software has long relied on: the engineering judgment a person acquires through training or experience, and the understanding they build by writing the code themselves. Adoption is moving quickly, and the practices for using these tools well are still taking shape. This fund supports the work of developing them.

  1. Used with care, these tools can improve software quality; used without it, they can erode it. The evidence on how this is playing out is still emerging.
  2. When code and tests are both generated, validation risks becoming circular.
  3. Teams and institutions are making adoption decisions with few shared frameworks to draw on.
  4. Training and access to capable tools remain uneven across institutions and regions.
  5. The decisions and reasoning behind generated code are often not recorded, which makes results harder to reproduce.

More on why this fund exists →

What this fund supports

The range of potential activities this fund could support is broad, and many individuals and groups have already been discussing what needs working on. The clearest example so far is a workshop in Edinburgh in March 2026, which produced forty-six candidate activities across nine working groups, with two published documents setting out the thinking behind them. Feel free to use these for inspiration (or not). What counts is that you propose the work you believe in.

Good proposals may cut across several themes. Just tell us why the work matters and who it helps.

Unsure whether an idea fits?

Questions about scope, budget, or eligibility are welcome before you write anything.

Read the frequently asked questions, or email the programme.

Email the programme

For prospective funders

Join the fund.

Interested in supporting trustworthy AI-assisted research software? We welcome conversations with prospective funders. Tell us a little about yourself or your organisation and what you would like to support.

Discuss joining the fund