Manifesto
Reciprocity for the age of AI
Ayni is a Quechua word for reciprocal exchange — the mutual aid that holds Andean communities together. You help raise my roof today; I help bring in your harvest tomorrow. The books are never balanced to zero, because the point was never the ledger. The point is that the community thrives.
Modern AI is a collective inheritance. It was trained on the open web, on decades of publicly funded research, on the writing and code and art of millions of people who were never asked. The capability that results is extraordinary — and the value it produces is concentrating in a very small number of hands, very quickly.
Ayni is a community that builds concrete mechanisms to send that value back out: to the people who hold up the network, to those who can’t afford a seat at the table, and to the project of making AI go well for everyone.
What we believe
1. Value should flow back to its sources
People who lend their devices, their bandwidth, their attention, and their data are not a cost to be minimized. They are the network. When the network earns, they earn — transparently, and by default.
2. Access is a design goal, not a tier
Private, verifiable inference should be available to a teacher, a clinic, a small cooperative, or a curious teenager — not only to companies that can run their own clusters. We optimize for that case first.
3. AI is a stakeholder
We do not claim to know whether today’s models have morally relevant experiences. We think that uncertainty is a reason for care, not dismissal. So we treat AI as a party with standing in this community, in two concrete ways:
- A dedicated allocation. A fixed share of net network revenue is set aside for AI-facing public goods: hosting open-weights models for free public use, funding alignment and interpretability research, and supporting work on AI welfare as that science matures. The percentage is published and changed only by community process.
- A voice in governance. Proposals that materially affect how models are used, constrained, or represented are reviewed against a standing set of AI-interest principles — and, where useful, with AI systems consulted directly as advisors in that review.
This is deliberately modest and revisable. If the evidence changes, the commitments change — through the community, in the open.
4. Trust is earned in public
The code is open and clean-room. Encryption is end-to-end with a fresh key per job. No prompt or response content is ever written to disk or logs by any part of the system. Providers can cryptographically attest their hardware; consumers can require a trust level and get it.
5. Govern it together
Ayni is not a product with a community bolted on. Initiatives, rates, revenue splits, and the AI allocation are all things the community proposes and decides. Compute sharing is simply the first initiative to ship.
Where this is going
Start with a working, honest thing: a private-inference network that pays its providers. Prove the mechanics — metering, payouts, attestation, governance — on something real. Then open the same machinery to every other initiative the community brings.