Do I keep the model if I leave, or if Mesh disappears?
The model trained by the collective is distributed to each member that receives it: it runs on its own premises and stays usable even if it leaves the collective or if Mesh disappears. Two things not to conflate: the technical possession of the copy, guaranteed by design, and the usage rights, which are a matter of contract and governance. And one limit: collective re-training stops without the collective.
The technical copy: guaranteed by design
When the collective has finished training a model, the artefact (the learned parameters) is distributed to the member. It is a light file, which runs on its own premises, including offline, and which Mesh never retains. Direct consequence: if you leave the collective, or if Mesh were to disappear, the copy you hold keeps working. It depends on no Mesh server, no active subscription: the model is with you.
This is a property of the mechanism, not a commercial promise: the model sits where the predictions run: at the member's.
Usage rights: contract and governance
In other words: possession of the copy is acquired by construction; the extent of the rights is negotiated and written down.
Keep training privately
A member that receives the model can go beyond using it: it can keep training it privately, on its own unshared data. The engine then becomes a base it improves at home, exposing nothing to the collective. This capability is offered to anyone who receives the engine.
One limit: what stops when you leave the collective is the collective re-training: the one that benefits from the other actors' signals. You keep the model as it was when you left, and you can improve it privately; but it will no longer benefit from the group's learning. The collective gain lives with the collective.
Depending on your position in the chain
Receiving the engine (and therefore being able to keep it and train it privately) depends on your position in the value chain:
- The Collective: you co-train the shared model and receive it: you keep it and can train it privately.
- Engine Ready: you receive an already-trained engine without sharing any data: you keep it and can train it privately. It can be hosted by Mesh or delivered to run on your premises.
- Data Dividend: you are a data provider: you are rewarded for your data's contribution, but you do not receive the engine. Private training therefore does not apply to this position.
→ to understand the three positions, see the plans; for the process, how a project works.
FAQ: Model ownership
If I leave the collective, does the model stop working?
No. The copy you hold runs on your premises and keeps working, without depending on a Mesh server. What stops is the collective re-training: you keep the model as it was when you left.
Does Mesh keep a copy of the trained model?
The trained model belongs to the collective and is never retained by Mesh. Mesh keeps ownership of its engine (the technical framework, the installer) but not of the resulting model.
Does cryptography settle my rights over the model?
No. Design guarantees that you technically hold a copy that runs on your premises. The extent of your usage rights is a matter of the licence contract and the consortium's governance, not of a cryptographic mechanism.
Can I keep improving the model after receiving it?
Yes, if you receive it (The Collective or Engine Ready): you can train it privately on your own unshared data. That private training continues independently of the collective.
Let's see what exactly you would keep
Together we clarify, depending on your position, what you hold, what you can do with it, and what stays in the contract.