QBC network · planned
Model-agnostic interoperability and external providers.
plannedEverything on this page is planned. None of it is operating today.
No partnership or integration with any AI provider exists. The systems named here are examples of what the network is designed to serve, nothing more.
The network is designed to serve any model; today it serves one. No external model is registered or supported. This page sets out what has to be built for that to change.
Who it is designed to serve.
Examples only. Naming a system here does not mean it uses, supports or has any relationship with the network.
- for example Claude
- for example OpenAI models
- for example Gemini
- for example Llama
- for example local models
The intent is that a developer using a hosted model, or running a local one, could record which model answered and when, without that model's maker having to take part and without any dependency on AethersMind.
How a provider integration would work.
The planned flow for the first demonstrator (milestone M1).
- 01An adapter sits beside the application's call to a model. It hashes the request and the response. The content itself does not go on chain.
- 02The adapter or the operator signs a receipt: model identifier, input hash, output hash, the signing key and an evidence-level tag.
- 03The network checks the signature, records the receipt and finalises it.
- 04Anyone with the receipt can run a verifier tool that checks it against chain state.
A receipt of this kind is evidence level L2: a named key vouching for hashes. It is a signed attestation, not independent verification. A working API integration is not a commercial partnership.
What has to be built.
Identity
plannedThe registry accepts submissions only from validators and keeps one lineage. Needed: a model namespace, registrant accounts with a deposit, a parent pointer per checkpoint, and hashes for the evaluation report and model card.
Limitation: Milestone M2.
Receipts
plannedNo receipt type exists. Needed: a signed receipt carrying the model identifier, input hash, output hash, provider key and an evidence-level tag.
Limitation: Milestone M1, M3.
Adapters
plannedNeeded: provider adapters that hash the request and response for a hosted provider API and for a local model, and a verifier command-line tool that checks a receipt against chain state.
Limitation: Milestone M1.
Settlement
plannedToday's settlement contracts assume one model. Needed: settlement keyed by model and registrant.
Limitation: Milestone M2, M3.
Proof checking
plannedNeeded: the network verifies a hardware attestation (L3) and a third-party zk proof (L4) and records the results.
Limitation: Milestone M4.
Trust
plannedUntil validators outside SusyLabs exist, any external model's record would be witnessed only by SusyLabs.
Limitation: Milestone M6.
Where AethersMind fits.
Aether / AethersMind is the reference implementation running on the network. It is never a dependency for other AI systems. The first demonstrator is required to have zero AethersMind code in its path, and AethersMind itself is to be migrated into the open registry as registrant 1, on the same terms as any other model.