ORÆ Open Intelligence Initiative — coordination and funding for the open AI commons.

Mission

The ORÆ Open Intelligence Initiative helps researchers, developers and maintainers transform open AI projects into fundable, auditable and reusable public infrastructure.

Open AI does not only need more models. It needs better coordination, long-term maintenance, transparent funding, reusable datasets, trustworthy evaluation tools and public memory. The Initiative is designed to coordinate these missing layers — not to replace open source projects, but to fund and connect the public-good infrastructure they depend on.

What we fund

Six layers of the open AI commons:

• Open ModelsFoundation and task-specific models with meaningful access to training method, weights or both, released under usable licenses.
• Open Datasets & MemoryReusable, well-documented datasets and long-term memory infrastructure with clear provenance and usable licensing.
• Evaluation, Safety & GovernanceBenchmarks, evaluation harnesses, safety tooling and governance frameworks the broader ecosystem can rely on.
• Open Agents & CoordinationAgent frameworks, protocols and coordination tools designed for interoperability rather than lock-in.
• Maintenance & Critical DependenciesTime-bound support for maintainers of the libraries, runtimes and infrastructure that hold the ecosystem together.
• Documentation & ReproducibilityDocumentation, tutorials, reproducible recipes and onboarding paths — the unsung infrastructure of a healthy commons.

ORÆ Grants — funding formats

Different projects need different shapes of support. The program is designed around four funding formats — each with its own scope, duration and review depth.

• Micro-grantsDocumentation, benchmarks, bug fixes, small datasets, evaluation tools — narrow scope, short timeline.
• Builder grantsOpen AI components with clear milestones and public deliverables — medium scope, structured roadmap.
• Maintainer fellowshipsTime-bound support for maintainers of critical AI infrastructure — sustains work that would otherwise stop.
• Strategic grantsLarger initiatives around models, datasets, evaluation, safety or governance — long horizon, public reporting.

Selection criteria

Selection is designed around public criteria, human review and transparent decision logs. No criterion alone is decisive — the review weighs them together.

• Public usefulnessDoes the work strengthen the open AI commons for many, not just for one team?
• Technical feasibilityIs the scope realistic given the team, budget and timeline?
• Openness levelOpen Source AI, Open Weights or Research Commons — clearly stated, not vague.
• License clarityLicenses are explicit and compatible with the openness level claimed.
• ReusabilityDesigned so others can adopt, extend or audit the work.
• Safety & risk awarenessKnown risks named and addressed; refusal zones respected.
• Maintenance planThe plan for keeping the work alive after the grant ends.
• Documentation qualityDocumentation, examples and onboarding are part of the deliverable, not afterthoughts.
• Ecosystem impactHow the work fits and strengthens what already exists in the commons.

Governance

ORÆ Open Intelligence Initiative is designed around public criteria, human review, conflict-of-interest declarations and transparent decision logs. Selection, disbursement and reporting are public-facing by default.

• Public selection criteriaStated up front, applied consistently, revisable in the open.
• Conflict-of-interest declarationsReviewers declare relationships before deliberation; recusals are logged.
• Transparent decision logsFunding decisions and their reasoning are summarised in the public log.
• Open deliverablesOutputs are released under the openness level the project claimed.
• Maintenance-first mindsetLong-term sustainability is treated as a primary deliverable, not an afterthought.
• Human reviewNo purely automated funding decisions. Humans accountable to humans.

ORÆ Atlas

Each project can be listed in the ORÆ Atlas — a public registry of the systems, datasets and tools the program coordinates. Atlas entries are designed to be machine-readable and citation-friendly.

FAQ