A nonprofit, community-governed effort. These are the collaborations we are open to — and the help that unblocks real work.
We do not sell the work, grant exclusive licenses, or raise venture capital. The compute, researcher time, and engineering behind an open AI-for-science stack are real costs the community absorbs. We collaborate with partners who want the open future of scientific discovery to win — on terms that keep the work public.
What you can give. Inference credits on frontier closed models (Anthropic, OpenAI, Google) or open-weights API providers (Together, Fireworks, Groq, DeepInfra, SiliconFlow). Six- to twelve-month grants, renewable; multi-year preferred. API keys bound to our org, not per-researcher reseller accounts.
What you get. Acknowledgment in every paper, evaluation report, and dataset that uses the credit. Provider logo on the relevant Pillar pages. Named mention in release notes of every artifact the credit produces.
What you can give. H100 / H200 / B200 / MI300 class accelerators with multi-node interconnect for anything above 7B; A100s and consumer-grade cards for LoRA adapters and evaluation simulation. 1–2k GPU-hours/month on H100-class unblocks a LoRA recipe; 10k GPU-hours unblocks a full continued-pretraining run. Bare-metal grants, research-credit vouchers (AWS / GCP / Azure / Oracle / Lambda / CoreWeave / Modal / Runpod / Together Train), or dedicated reserved capacity at research rate — all work.
What you get. Provider named in the recipe file, the training log, the model card, and the paper. Reproducibility artifacts cite the exact partition.
What you can give. Unrestricted grants, project-specific grants, recurring donations, or fiscal sponsorship. Foundations: multi-year grants on open infrastructure programs. Corporations: sponsored research on open-by-default terms. Individuals: recurring donations via fiscal sponsor (Open Collective, NumFOCUS). US-based tax-deductible giving is possible today.
What you get. Named in the annual report, on the Pillar page your grant funds, and in the acknowledgments of resulting papers. Anonymous donations also supported.
What you can give. Co-authored papers, research-group partnerships, shared benchmarks, student rotations, or lab exchanges. Concretely: authoring 5–20 skills in your subdomain over a semester, contributing a task bundle to Evaluation with replayable ground truth, hosting a student to build a subdomain folder, or running the Agent harness on a real problem and publishing the artifact.
What you get. Free access to the Agent harness, Evaluation environment, and Skills registry. Engineering support to integrate your lab’s tools. Named co-authorship on any paper that uses your contribution. A committed reviewer inside the collective for your subdomain.
What you can give. Coverage, interviews, newsletter syndication, podcast episodes that go technical, conference talks, or translation into Chinese, Japanese, Korean, French, Spanish and other contributor-base languages. Profiles of a Pillar, bilingual translations of architecture pages, regional editors pitching science outlets — all qualify.
What you get. Unfettered access to contributors and technical material. No PR-filtering, no embargo games. Source files, datasets, and figures under open license for any piece that runs about us.
What you can give. Pro-bono nonprofit counsel and IP review, community management (Discord / forum moderation, onboarding), event logistics (sprints, workshops, conference meetups), translations, documentation writers, accessibility reviewers, or designers who care about editorial typography on the web.
What you get. Real responsibility (not box-checking volunteer work), named credit on the governance page, and a seat in the decision-making process for the part of the collective you are sustaining.
Or email [email protected] directly.