Limitations & FAQ
#Status
Is this on mainnet? No. Everything runs on Bittensor testnet, subnet 566. There is no mainnet registration, no mainnet wallet and no announced date. Any claim otherwise is not from Everyframe.
Can I just register and start mining? No. Enrolment is by operator invitation after your hotkey ownership is verified and your VM's measurements are reviewed. There is no self-registration endpoint and no trust-on-first-use.
Is Everyframe Studio traffic flowing to miners now? Production stays Direct — Studio calls providers itself — and routing to the subnet is enabled per pilot window. A model in the catalogue is not a promise of work.
#Money
Do I get paid in USD? Not automatically. earnings shows serving-fee accounting (review-held / approved / disputed). USD settlement and wallet payouts are not implemented in the pilot.
Do I get paid in TAO? Approved work sets weights on testnet 566, which pays testnet alpha — a test token with no redeemable value. The mechanism has been verified end to end (2 jobs : 1 job → 2 : 1 alpha), but the pilot is about proving the mechanism, not income.
What do I pay? Phala hosting (≈ $0.058/h for tdx.small plus storage) and every render on your Fal key at the provider's price. Reward targets currently equal the Fal reference, so there is no margin built in.
Why does my Fal bill differ from the reference estimate? The reference is a public list-price estimate. Actual token usage, provider promotions, failed requests and billing quirks all differ. Everyframe never invents an "actual" provider cost.
#Hardware and hosting
Do I need a GPU? No. The worker is a 1 vCPU / 2 GiB confidential VM that talks to a provider API. Generation happens at the provider.
Can I run the worker on my own TDX machine? Not in this release. The invitation pins Phala Cloud, the dstack-0.5.9 guest and Phala's KMS. Phala is a trusted KMS operator in the pilot's threat model.
Can I run it without a TEE, on a normal VPS? No. Quote verification is mandatory and fails closed; there is no simulation mode in production entry points.
Why is the image private? The pilot image lives in a private GHCR repository and is pulled with scoped read-only credentials issued by the operator. A public reviewed image or a revised boot workflow is a prerequisite for wider onboarding.
#Operating
doctor says not ready — what now? Read the failing check. Before deploy it is normal. After deploy, it usually means admission review is pending on the operator's side; after activate, it means the post-restart admission has not landed yet.
A command was interrupted / my terminal died. Run everycli miner reconcile. It is read-only and tells you whether the VM, update, start or stop actually completed. Do not delete state or create a second VM.
Can I change the coordinator URL, disable TLS checks or add debug logging? No. Those flags do not exist and the corresponding environment variables make the CLI refuse to start.
Can the operator read my prompts or take my keys? Everyframe's coordinator sees the prompts and the videos — it is the job source. Your provider key is encrypted locally to the Phala app key and released only into the admitted enclave; the operator never sees it. The worker has no shell, no HTTP server and no generic signing endpoint.
Windows? Not supported in this release (file-permission semantics).
#What is deliberately missing
- Permissionless enrolment and local wallet signing
- Automatic payouts, on-chain earnings queries and USD settlement
- Keychain-encrypted local storage
- Automatic hosting expiry or a VM delete command
- Multi-validator consensus, replicated coordinators, independent security or KMS audits
- Live-tested
updateandstartpaths on real cloud (unit- and integration-tested only) - 34 of 37 registered models have not been paid- or TEE-live tested
- Mainnet
If you need something on this list, ask — the pilot's roadmap is set by what miners and customers actually hit.