Sponsor our research
Octopus Research Institute is the knowledge entity in the Octopus ecosystem — separate from Octopus Core, which builds infrastructure, and Octopus Foundation, which serves the community. We publish open, applied and responsible AI research: work on sovereign and local inference, cross-chip validation, reproducible evaluation with full provenance, governed agent systems, and community-governed Auslan accessibility. We publish negative results and limitations, version and correct what we release, and never fabricate — which is precisely the kind of research that grant-free open work struggles to sustain.
Why this research needs support
Honest, reproducible research has a cost structure that hype does not. Proving that a runtime's numbers match a reference across Apple silicon, ROCm and CUDA means running the same models many times, on real hardware, and keeping the audit trail. Sovereign and local inference — running models you control, across chips, with byte-identical failover — has to be demonstrated, not asserted. Reproducible evaluation with full provenance means storing the evidence, not just the headline. None of this is free, and it does not sell a product on its own.
Publishing what did not work is a public good that no customer pays for. A negative result, a corrected version, a documented limitation, or the open tooling that lets others reproduce a claim — these are the connective tissue of trustworthy AI, yet they sit outside the usual funding logic. The community-governed Auslan and Deaf-accessibility direction is deliberately gated on community leadership and consent, which means it moves at the pace of trust rather than the pace of a grant cycle. Support is what lets us keep this work open, reproducible and honest instead of quietly dropping the parts that are hard to fund.
Four ways to support the work
- Government & public sector
Independent, auditable evidence for sovereign AI
If your concern is keeping data and models under national or organisational control, our work on sovereign and local inference gives you evidence rather than vendor claims: models you run yourself, across chips, with byte-identical failover and an audit trail. The community-governed Auslan direction extends the same public-interest lens to Deaf accessibility. Supporting the Institute funds independent, auditable, public-good research whose findings belong in the open, not behind a procurement pitch.
- Hardware & platform vendors
Independent, reproducible cross-chip validation
Chip, accelerator and platform makers benefit from third-party evidence that their hardware runs real models correctly — validation that a runtime's numbers match a reference across Apple silicon, ROCm and CUDA, done independently and reproducibly. The most useful way to help is in kind: hardware and compute access so we can run these experiments on your platform and publish the method and results in full. What you fund is a reproducible validation, not a favourable review — the findings stand as measured.
- Research & industry partners
Joint research and open, co-authored outputs
Partners can collaborate directly on the research: joint studies, dataset evaluation, benchmark contribution, and co-authored open outputs on governed agent systems, reproducible evaluation, or cross-chip inference. We work in the open and version our claims, so a partnership produces a durable, citable record rather than a one-off deck. Bring a problem, a dataset, or a benchmark, and we will design the experiment so the result is reproducible by anyone.
- Philanthropic & community
Fund a programme, or the foundations under all of it
You can fund a specific programme, or the reproducibility and open-tools work that underpins everything the Institute does. The direction closest to community is the community-governed Auslan and Deaf-accessibility mission, which advances only under Deaf community leadership and consent — support here helps sustain that governance, not rush it. Philanthropic backing keeps the unglamorous, essential parts alive: the open tools, the evidence infrastructure, and the honest negative results.
What your support makes possible
Compute for reproducible experiments
Running the same models many times, across Apple silicon, ROCm and CUDA, to prove a runtime's numbers match a reference — and to keep byte-identical failover and audit trails honest rather than aspirational.
Maintenance of open tools
Keeping the open tooling that lets others reproduce our claims working, documented and current, so a published result can be checked by anyone rather than taken on trust.
Community governance for Auslan
Sustaining the Deaf-community leadership, consent processes and slow, careful pace that the Auslan multimodal accessibility direction is deliberately gated on.
Publishing & evidence infrastructure
Storing provenance, versions and corrections so evaluations carry their full evidence — including the negative results and limitations most publishing quietly omits.
Support does not buy influence over findings
Independence is not a courtesy we extend to sponsors; it is the whole value of the work. Funding and in-kind contributions never buy influence over what we find or how we report it. Contributions — including hardware or compute from a platform vendor — are disclosed on the relevant research pages, so a reader can see who supported a piece of work and judge it accordingly. Results are never altered to suit a sponsor: a negative result stays negative, a limitation stays documented, and a correction is published when we are wrong. The Institute does not become a marketing arm of any sponsor, and no contribution grants a right of review, approval, or veto over our conclusions. If honest findings and a sponsorship ever come into conflict, the findings win.
Questions sponsors ask
- Does sponsoring change what you publish?
- No. Support never buys influence over findings or how they are reported. Negative results, limitations and corrections stand as measured, and no contribution grants any right of review, approval or veto. Your contribution is disclosed on the relevant research pages so readers can weigh it for themselves.
- Can we fund just one programme?
- Yes. You can support a specific programme — for example cross-chip validation or the community-governed Auslan work — or the reproducibility and open-tools foundations that underpin everything. For the Auslan direction, note it advances only under Deaf community leadership and consent, so support sustains that governance rather than accelerating it.
- How is our support acknowledged?
- Funding and in-kind contributions are disclosed on the research pages they support, so acknowledgement is tied to the specific work rather than a blanket endorsement. What we will not do is present the Institute or its findings as a marketing channel for a sponsor — acknowledgement records who helped, it does not shape conclusions.
- Do you take payments on this website?
- No. This site is informational and does not process donations or payments of any kind. Sponsorship and in-kind support are arranged directly by contacting the Institute, so we can scope the work, agree how contributions are disclosed, and confirm any community-governance requirements before anything begins.